rpms/qt5-qtwebengine
3
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
968c8c053a
qt5-qtwebengine/0004-qtwebengine-opensource-src-5.9.0-no-sse2.patch

4972 lines
180 KiB
Diff
Raw Normal View History

Revert "- Fix icu and nspr patches. One to go" This reverts commit ee046f9af5635bb49d85121ec73e3cfe9415105a. I already told you last time (5.8) that this way of maintaining patches (importing them all into git and reexporting them from there each time) is not how I work (I want only those patches rebased and renamed that actually need rebasing) and that you are only causing me more work. Please keep your hands off my package from now on.
2017-06-05 21:59:21 +00:00
From 844246bd10971ec348f044bc7a13acb17b9ce033 Mon Sep 17 00:00:00 2001
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
From: Helio Chissini de Castro <helio@kde.org>
Date: Wed, 31 May 2017 14:38:20 +0200
Subject: [PATCH] qtwebengine-opensource-src-5.9.0-no-sse2
---
.../chromium/breakpad/src/build/common.gypi | 8 +-
src/3rdparty/chromium/cc/BUILD.gn | 39 +-
src/3rdparty/chromium/media/BUILD.gn | 20 +
src/3rdparty/chromium/media/base/BUILD.gn | 48 +-
src/3rdparty/chromium/media/base/media.cc | 4 +
.../chromium/media/base/simd/convert_yuv_to_rgb.h | 34 +
.../media/base/simd/convert_yuv_to_rgb_mmx.asm | 24 +
.../media/base/simd/convert_yuv_to_rgb_x86.cc | 56 +
src/3rdparty/chromium/media/base/simd/filter_yuv.h | 6 +
.../chromium/media/base/simd/filter_yuv_mmx.cc | 79 +
.../chromium/media/base/simd/sinc_resampler_sse.cc | 50 +
.../chromium/media/base/simd/vector_math_sse.cc | 118 +
src/3rdparty/chromium/media/base/sinc_resampler.cc | 81 +-
src/3rdparty/chromium/media/base/sinc_resampler.h | 4 +
.../chromium/media/base/sinc_resampler_perftest.cc | 4 +
.../chromium/media/base/sinc_resampler_unittest.cc | 5 +
src/3rdparty/chromium/media/base/vector_math.cc | 150 +-
src/3rdparty/chromium/media/base/vector_math.h | 5 +
.../chromium/media/base/vector_math_perftest.cc | 35 +-
.../chromium/media/base/vector_math_testing.h | 2 +-
.../chromium/media/base/vector_math_unittest.cc | 4 +
src/3rdparty/chromium/media/base/yuv_convert.cc | 50 +-
.../chromium/media/base/yuv_convert_perftest.cc | 79 +-
.../chromium/media/base/yuv_convert_unittest.cc | 159 +-
src/3rdparty/chromium/skia/BUILD.gn | 52 +-
src/3rdparty/chromium/skia/ext/convolver.cc | 11 +-
src/3rdparty/chromium/skia/ext/convolver.h | 1 +
.../Source/modules/webaudio/AudioParamTimeline.cpp | 8 +-
.../Source/platform/audio/DirectConvolver.cpp | 40 +-
.../WebKit/Source/platform/audio/DirectConvolver.h | 7 +
.../WebKit/Source/platform/audio/SincResampler.cpp | 31 +-
.../WebKit/Source/platform/audio/SincResampler.h | 7 +
.../WebKit/Source/platform/audio/VectorMath.cpp | 97 +-
.../WebKit/Source/platform/audio/VectorMath.h | 23 +
.../graphics/cpu/x86/WebGLImageConversionSSE.h | 2 +-
.../platform/graphics/gpu/WebGLImageConversion.cpp | 12 +-
src/3rdparty/chromium/third_party/qcms/BUILD.gn | 17 +-
.../webrtc/common_audio/real_fourier.cc | 9 +
src/3rdparty/chromium/v8/BUILD.gn | 72 +-
src/3rdparty/chromium/v8/src/make-v8-sse2-gyp.sh | 56 +
src/3rdparty/chromium/v8/src/v8.gyp | 12 +-
src/3rdparty/chromium/v8/src/v8_sse2.gyp | 2313 ++++++++++++++++++++
src/core/core_module.pro | 30 +-
43 files changed, 3537 insertions(+), 327 deletions(-)
create mode 100644 src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb_mmx.asm
create mode 100644 src/3rdparty/chromium/media/base/simd/filter_yuv_mmx.cc
create mode 100644 src/3rdparty/chromium/media/base/simd/sinc_resampler_sse.cc
create mode 100644 src/3rdparty/chromium/media/base/simd/vector_math_sse.cc
create mode 100644 src/3rdparty/chromium/v8/src/make-v8-sse2-gyp.sh
create mode 100644 src/3rdparty/chromium/v8/src/v8_sse2.gyp
diff --git a/src/3rdparty/chromium/breakpad/src/build/common.gypi b/src/3rdparty/chromium/breakpad/src/build/common.gypi
index b9466a325..8ad78e24f 100644
--- a/src/3rdparty/chromium/breakpad/src/build/common.gypi
+++ b/src/3rdparty/chromium/breakpad/src/build/common.gypi
no-sse2.patch: Fix/simplify disable_sse2 logic in breakpad/src/build/common.gypi That file is not really used in QtWebEngine anyway, but simply defaulting disable_sse2 to 1 is the right thing. The variable is ignored on everything other than ia32 (i686) anyway, even on x64 (x86_64). And setting conditional defaults might not work as expected in gyp.
2016-01-17 05:32:37 +00:00
@@ -80,8 +80,8 @@
# The system root for cross-compiles. Default: none.
'sysroot%': '',
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
no-sse2.patch: Fix/simplify disable_sse2 logic in breakpad/src/build/common.gypi That file is not really used in QtWebEngine anyway, but simply defaulting disable_sse2 to 1 is the right thing. The variable is ignored on everything other than ia32 (i686) anyway, even on x64 (x86_64). And setting conditional defaults might not work as expected in gyp.
2016-01-17 05:32:37 +00:00
- # On Linux, we build with sse2 for Chromium builds.
- 'disable_sse2%': 0,
+ # Do not assume SSE2 by default (Fedora patch).
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+ 'disable_sse2%': 1,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
},
'target_arch%': '<(target_arch)',
no-sse2.patch: Fix/simplify disable_sse2 logic in breakpad/src/build/common.gypi That file is not really used in QtWebEngine anyway, but simply defaulting disable_sse2 to 1 is the right thing. The variable is ignored on everything other than ia32 (i686) anyway, even on x64 (x86_64). And setting conditional defaults might not work as expected in gyp.
2016-01-17 05:32:37 +00:00
@@ -725,17 +725,13 @@
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
'conditions': [
['disable_sse2==0', {
'cflags': [
- '-march=pentium4',
'-msse2',
'-mfpmath=sse',
],
}],
],
- # -mmmx allows mmintrin.h to be used for mmx intrinsics.
- # video playback is mmx and sse2 optimized.
'cflags': [
'-m32',
- '-mmmx',
],
'ldflags': [
'-m32',
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/cc/BUILD.gn b/src/3rdparty/chromium/cc/BUILD.gn
index e461a898d..b50659b89 100644
--- a/src/3rdparty/chromium/cc/BUILD.gn
+++ b/src/3rdparty/chromium/cc/BUILD.gn
@@ -567,13 +567,6 @@ component("cc") {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
"trees/tree_synchronizer.h",
]
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
- if (current_cpu == "x86" || current_cpu == "x64") {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
- sources += [
- "raster/texture_compressor_etc1_sse.cc",
- "raster/texture_compressor_etc1_sse.h",
- ]
- }
-
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
configs += [ "//build/config:precompiled_headers" ]
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
public_deps = [
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -583,6 +576,7 @@ component("cc") {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
deps = [
"//base",
"//base/third_party/dynamic_annotations",
+ "//cc:cc_opts",
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
"//cc/proto",
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
"//cc/surfaces:surface_id",
"//gpu",
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -612,6 +606,37 @@ component("cc") {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
}
}
+source_set("cc_opts") {
+ public_deps = [
+ "//cc:cc_opts_sse",
+ ]
+}
+
+source_set("cc_opts_sse") {
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ if (current_cpu == "x86" || current_cpu == "x64") {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ deps = [
+ "//base",
+ ]
+
+ defines = [ "CC_IMPLEMENTATION=1" ]
+
+ if (!is_debug && (is_win || is_android)) {
+ configs -= [ "//build/config/compiler:optimize" ]
+ configs += [ "//build/config/compiler:optimize_max" ]
+ }
+
+ sources = [
+ "raster/texture_compressor.h",
+ "raster/texture_compressor_etc1.h",
+ "raster/texture_compressor_etc1_sse.cc",
+ "raster/texture_compressor_etc1_sse.h",
+ ]
+
+ cflags = [ "-msse2" ]
+ }
+}
+
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+
static_library("test_support") {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
testonly = true
sources = [
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/BUILD.gn b/src/3rdparty/chromium/media/BUILD.gn
index 652aad4d9..063254830 100644
--- a/src/3rdparty/chromium/media/BUILD.gn
+++ b/src/3rdparty/chromium/media/BUILD.gn
@@ -832,6 +832,26 @@ component("shared_memory_support") {
"//base",
"//ui/gfx/geometry",
]
+ if (current_cpu == "x86" || current_cpu == "x64") {
+ deps += [
+ ":shared_memory_support_sse",
+ ]
+ }
+}
+
+if (current_cpu == "x86" || current_cpu == "x64") {
+ source_set("shared_memory_support_sse") {
+ sources = [
+ "base/simd/vector_math_sse.cc",
+ ]
+ configs += [
+ "//media:media_config",
+ "//media:media_implementation",
+ ]
+ if (!is_win) {
+ cflags = [ "-msse" ]
+ }
+ }
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
}
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
# TODO(watk): Refactor tests that could be made to run on Android. See
diff --git a/src/3rdparty/chromium/media/base/BUILD.gn b/src/3rdparty/chromium/media/base/BUILD.gn
index d288ec787..fac712c00 100644
--- a/src/3rdparty/chromium/media/base/BUILD.gn
+++ b/src/3rdparty/chromium/media/base/BUILD.gn
@@ -336,11 +336,12 @@ target(link_target_type, "base") {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
}
if (current_cpu == "x86" || current_cpu == "x64") {
- sources += [
- "simd/convert_rgb_to_yuv_sse2.cc",
- "simd/convert_rgb_to_yuv_ssse3.cc",
- "simd/convert_yuv_to_rgb_x86.cc",
- "simd/filter_yuv_sse2.cc",
+ sources += [ "simd/convert_yuv_to_rgb_x86.cc" ]
+ deps += [
+ ":media_yasm",
+ ":media_mmx",
+ ":media_sse",
+ ":media_sse2",
]
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
deps += [ ":media_yasm" ]
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
}
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -539,10 +540,47 @@ source_set("perftests") {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
}
if (current_cpu == "x86" || current_cpu == "x64") {
+ source_set("media_mmx") {
+ sources = [ "simd/filter_yuv_mmx.cc" ]
+ configs += [ "//media:media_config" ]
+ if (!is_win) {
+ cflags = [ "-mmmx" ]
+ }
+ }
+
+ source_set("media_sse") {
+ sources = [
+ "simd/sinc_resampler_sse.cc",
+ ]
+ configs += [
+ "//media:media_config",
+ "//media:media_implementation",
+ ]
+ if (!is_win) {
+ cflags = [ "-msse" ]
+ }
+ }
+
+ source_set("media_sse2") {
+ sources = [
+ "simd/convert_rgb_to_yuv_sse2.cc",
+ "simd/convert_rgb_to_yuv_ssse3.cc",
+ "simd/filter_yuv_sse2.cc",
+ ]
+ configs += [
+ "//media:media_config",
+ "//media:media_implementation",
+ ]
+ if (!is_win) {
+ cflags = [ "-msse2" ]
+ }
+ }
+
import("//third_party/yasm/yasm_assemble.gni")
yasm_assemble("media_yasm") {
sources = [
"simd/convert_rgb_to_yuv_ssse3.asm",
+ "simd/convert_yuv_to_rgb_mmx.asm",
"simd/convert_yuv_to_rgb_sse.asm",
"simd/convert_yuva_to_argb_mmx.asm",
"simd/empty_register_state_mmx.asm",
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/media.cc b/src/3rdparty/chromium/media/base/media.cc
index 38daf05b8..bb43b405f 100644
--- a/src/3rdparty/chromium/media/base/media.cc
+++ b/src/3rdparty/chromium/media/base/media.cc
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
@@ -10,6 +10,8 @@
#include "base/metrics/field_trial.h"
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "base/trace_event/trace_event.h"
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
#include "media/base/media_switches.h"
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+#include "media/base/sinc_resampler.h"
+#include "media/base/vector_math.h"
#include "media/base/yuv_convert.h"
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
#if defined(OS_ANDROID)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -40,6 +42,8 @@ class MediaInitializer {
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
TRACE_EVENT_WARMUP_CATEGORY("media");
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
// Perform initialization of libraries which require runtime CPU detection.
+ vector_math::Initialize();
+ SincResampler::InitializeCPUSpecificFeatures();
InitializeCPUSpecificYUVConversions();
#if !defined(MEDIA_DISABLE_FFMPEG)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb.h b/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb.h
index a421ff566..8373fde90 100644
--- a/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb.h
+++ b/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb.h
@@ -65,6 +65,17 @@ MEDIA_EXPORT void ConvertYUVToRGB32_SSE(const uint8_t* yplane,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
int rgbstride,
YUVType yuv_type);
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+MEDIA_EXPORT void ConvertYUVToRGB32_MMX(const uint8_t* yplane,
+ const uint8_t* uplane,
+ const uint8_t* vplane,
+ uint8_t* rgbframe,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ int width,
+ int height,
+ int ystride,
+ int uvstride,
+ int rgbstride,
+ YUVType yuv_type);
+
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
MEDIA_EXPORT void ConvertYUVAToARGB_MMX(const uint8_t* yplane,
const uint8_t* uplane,
const uint8_t* vplane,
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -86,6 +97,13 @@ MEDIA_EXPORT void ScaleYUVToRGB32Row_C(const uint8_t* y_buf,
ptrdiff_t source_dx,
const int16_t* convert_table);
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+MEDIA_EXPORT void ConvertYUVToRGB32Row_MMX(const uint8_t* yplane,
+ const uint8_t* uplane,
+ const uint8_t* vplane,
+ uint8_t* rgbframe,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ ptrdiff_t width,
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ const int16_t* convert_table);
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
MEDIA_EXPORT void LinearScaleYUVToRGB32Row_C(const uint8_t* y_buf,
const uint8_t* u_buf,
const uint8_t* v_buf,
@@ -94,6 +112,14 @@ MEDIA_EXPORT void LinearScaleYUVToRGB32Row_C(const uint8_t* y_buf,
ptrdiff_t source_dx,
const int16_t* convert_table);
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+MEDIA_EXPORT void ScaleYUVToRGB32Row_MMX(const uint8_t* y_buf,
+ const uint8_t* u_buf,
+ const uint8_t* v_buf,
+ uint8_t* rgb_buf,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ ptrdiff_t width,
+ ptrdiff_t source_dx,
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ const int16_t* convert_table);
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
MEDIA_EXPORT void LinearScaleYUVToRGB32RowWithRange_C(
const uint8_t* y_buf,
const uint8_t* u_buf,
@@ -104,6 +130,14 @@ MEDIA_EXPORT void LinearScaleYUVToRGB32RowWithRange_C(
int source_dx,
const int16_t* convert_table);
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+MEDIA_EXPORT void LinearScaleYUVToRGB32Row_MMX(const uint8_t* y_buf,
+ const uint8_t* u_buf,
+ const uint8_t* v_buf,
+ uint8_t* rgb_buf,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ ptrdiff_t width,
+ ptrdiff_t source_dx,
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ const int16_t* convert_table);
+
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
} // namespace media
// Assembly functions are declared without namespace.
diff --git a/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb_mmx.asm b/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb_mmx.asm
new file mode 100644
index 000000000..3650c9f73
--- /dev/null
+++ b/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb_mmx.asm
@@ -0,0 +1,24 @@
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+; Copyright (c) 2011 The Chromium Authors. All rights reserved.
+; Use of this source code is governed by a BSD-style license that can be
+; found in the LICENSE file.
+
+%include "third_party/x86inc/x86inc.asm"
+
+;
+; This file uses MMX instructions.
+;
+ SECTION_TEXT
+ CPU MMX
+
+; Use movq to save the output.
+%define MOVQ movq
+
+; extern "C" void ConvertYUVToRGB32Row_MMX(const uint8* y_buf,
+; const uint8* u_buf,
+; const uint8* v_buf,
+; uint8* rgb_buf,
+; ptrdiff_t width,
+; const int16* convert_table);
+%define SYMBOL ConvertYUVToRGB32Row_MMX
+%include "convert_yuv_to_rgb_mmx.inc"
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+
diff --git a/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb_x86.cc b/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb_x86.cc
index a06799bfd..819bb60a8 100644
--- a/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb_x86.cc
+++ b/src/3rdparty/chromium/media/base/simd/convert_yuv_to_rgb_x86.cc
@@ -47,6 +47,62 @@ void ConvertYUVAToARGB_MMX(const uint8_t* yplane,
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
EmptyRegisterState();
}
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+void ConvertYUVToRGB32_MMX(const uint8_t* yplane,
+ const uint8_t* uplane,
+ const uint8_t* vplane,
+ uint8_t* rgbframe,
+ int width,
+ int height,
+ int ystride,
+ int uvstride,
+ int rgbstride,
+ YUVType yuv_type) {
+ unsigned int y_shift = GetVerticalShift(yuv_type);
+ for (int y = 0; y < height; ++y) {
+ uint8_t* rgb_row = rgbframe + y * rgbstride;
+ const uint8_t* y_ptr = yplane + y * ystride;
+ const uint8_t* u_ptr = uplane + (y >> y_shift) * uvstride;
+ const uint8_t* v_ptr = vplane + (y >> y_shift) * uvstride;
+
+ ConvertYUVToRGB32Row_MMX(y_ptr,
+ u_ptr,
+ v_ptr,
+ rgb_row,
+ width,
+ GetLookupTable(yuv_type));
+ }
+
+ EmptyRegisterState();
+}
+
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+void ConvertYUVToRGB32_MMX(const uint8_t* yplane,
+ const uint8_t* uplane,
+ const uint8_t* vplane,
+ uint8_t* rgbframe,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ int width,
+ int height,
+ int ystride,
+ int uvstride,
+ int rgbstride,
+ YUVType yuv_type) {
+ unsigned int y_shift = GetVerticalShift(yuv_type);
+ for (int y = 0; y < height; ++y) {
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ uint8_t* rgb_row = rgbframe + y * rgbstride;
+ const uint8_t* y_ptr = yplane + y * ystride;
+ const uint8_t* u_ptr = uplane + (y >> y_shift) * uvstride;
+ const uint8_t* v_ptr = vplane + (y >> y_shift) * uvstride;
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+
+ ConvertYUVToRGB32Row_MMX(y_ptr,
+ u_ptr,
+ v_ptr,
+ rgb_row,
+ width,
+ GetLookupTable(yuv_type));
+ }
+
+ EmptyRegisterState();
+}
+
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
void ConvertYUVToRGB32_SSE(const uint8_t* yplane,
const uint8_t* uplane,
const uint8_t* vplane,
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/simd/filter_yuv.h b/src/3rdparty/chromium/media/base/simd/filter_yuv.h
index af30bd133..206fb7f11 100644
--- a/src/3rdparty/chromium/media/base/simd/filter_yuv.h
+++ b/src/3rdparty/chromium/media/base/simd/filter_yuv.h
@@ -20,6 +20,12 @@ MEDIA_EXPORT void FilterYUVRows_C(uint8_t* ybuf,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
int source_width,
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
uint8_t source_y_fraction);
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+MEDIA_EXPORT void FilterYUVRows_MMX(uint8_t* ybuf,
+ const uint8_t* y0_ptr,
+ const uint8_t* y1_ptr,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ int source_width,
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ uint8_t source_y_fraction);
+
MEDIA_EXPORT void FilterYUVRows_SSE2(uint8_t* ybuf,
const uint8_t* y0_ptr,
const uint8_t* y1_ptr,
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/simd/filter_yuv_mmx.cc b/src/3rdparty/chromium/media/base/simd/filter_yuv_mmx.cc
new file mode 100644
index 000000000..416aca699
--- /dev/null
+++ b/src/3rdparty/chromium/media/base/simd/filter_yuv_mmx.cc
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
@@ -0,0 +1,79 @@
+// Copyright (c) 2011 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#if defined(_MSC_VER)
+#include <intrin.h>
+#else
+#include <mmintrin.h>
+#endif
+
+#include "build/build_config.h"
+#include "media/base/simd/filter_yuv.h"
+
+namespace media {
+
+#if defined(COMPILER_MSVC)
+// Warning 4799 is about calling emms before the function exits.
+// We calls emms in a frame level so suppress this warning.
+#pragma warning(push)
+#pragma warning(disable: 4799)
+#endif
+
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+void FilterYUVRows_MMX(uint8_t* dest,
+ const uint8_t* src0,
+ const uint8_t* src1,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ int width,
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ uint8_t fraction) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ int pixel = 0;
+
+ // Process the unaligned bytes first.
+ int unaligned_width =
+ (8 - (reinterpret_cast<uintptr_t>(dest) & 7)) & 7;
+ while (pixel < width && pixel < unaligned_width) {
+ dest[pixel] = (src0[pixel] * (256 - fraction) +
+ src1[pixel] * fraction) >> 8;
+ ++pixel;
+ }
+
+ __m64 zero = _mm_setzero_si64();
+ __m64 src1_fraction = _mm_set1_pi16(fraction);
+ __m64 src0_fraction = _mm_set1_pi16(256 - fraction);
+ const __m64* src0_64 = reinterpret_cast<const __m64*>(src0 + pixel);
+ const __m64* src1_64 = reinterpret_cast<const __m64*>(src1 + pixel);
+ __m64* dest64 = reinterpret_cast<__m64*>(dest + pixel);
+ __m64* end64 = reinterpret_cast<__m64*>(
+ reinterpret_cast<uintptr_t>(dest + width) & ~7);
+
+ while (dest64 < end64) {
+ __m64 src0 = *src0_64++;
+ __m64 src1 = *src1_64++;
+ __m64 src2 = _mm_unpackhi_pi8(src0, zero);
+ __m64 src3 = _mm_unpackhi_pi8(src1, zero);
+ src0 = _mm_unpacklo_pi8(src0, zero);
+ src1 = _mm_unpacklo_pi8(src1, zero);
+ src0 = _mm_mullo_pi16(src0, src0_fraction);
+ src1 = _mm_mullo_pi16(src1, src1_fraction);
+ src2 = _mm_mullo_pi16(src2, src0_fraction);
+ src3 = _mm_mullo_pi16(src3, src1_fraction);
+ src0 = _mm_add_pi16(src0, src1);
+ src2 = _mm_add_pi16(src2, src3);
+ src0 = _mm_srli_pi16(src0, 8);
+ src2 = _mm_srli_pi16(src2, 8);
+ src0 = _mm_packs_pu16(src0, src2);
+ *dest64++ = src0;
+ pixel += 8;
+ }
+
+ while (pixel < width) {
+ dest[pixel] = (src0[pixel] * (256 - fraction) +
+ src1[pixel] * fraction) >> 8;
+ ++pixel;
+ }
+}
+
+#if defined(COMPILER_MSVC)
+#pragma warning(pop)
+#endif
+
+} // namespace media
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/simd/sinc_resampler_sse.cc b/src/3rdparty/chromium/media/base/simd/sinc_resampler_sse.cc
new file mode 100644
index 000000000..c5d03dc48
--- /dev/null
+++ b/src/3rdparty/chromium/media/base/simd/sinc_resampler_sse.cc
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
@@ -0,0 +1,50 @@
+// Copyright 2013 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "media/base/sinc_resampler.h"
+
+#include <xmmintrin.h>
+
+namespace media {
+
+float SincResampler::Convolve_SSE(const float* input_ptr, const float* k1,
+ const float* k2,
+ double kernel_interpolation_factor) {
+ __m128 m_input;
+ __m128 m_sums1 = _mm_setzero_ps();
+ __m128 m_sums2 = _mm_setzero_ps();
+
+ // Based on |input_ptr| alignment, we need to use loadu or load. Unrolling
+ // these loops hurt performance in local testing.
+ if (reinterpret_cast<uintptr_t>(input_ptr) & 0x0F) {
+ for (int i = 0; i < kKernelSize; i += 4) {
+ m_input = _mm_loadu_ps(input_ptr + i);
+ m_sums1 = _mm_add_ps(m_sums1, _mm_mul_ps(m_input, _mm_load_ps(k1 + i)));
+ m_sums2 = _mm_add_ps(m_sums2, _mm_mul_ps(m_input, _mm_load_ps(k2 + i)));
+ }
+ } else {
+ for (int i = 0; i < kKernelSize; i += 4) {
+ m_input = _mm_load_ps(input_ptr + i);
+ m_sums1 = _mm_add_ps(m_sums1, _mm_mul_ps(m_input, _mm_load_ps(k1 + i)));
+ m_sums2 = _mm_add_ps(m_sums2, _mm_mul_ps(m_input, _mm_load_ps(k2 + i)));
+ }
+ }
+
+ // Linearly interpolate the two "convolutions".
+ m_sums1 = _mm_mul_ps(m_sums1, _mm_set_ps1(
+ static_cast<float>(1.0 - kernel_interpolation_factor)));
+ m_sums2 = _mm_mul_ps(m_sums2, _mm_set_ps1(
+ static_cast<float>(kernel_interpolation_factor)));
+ m_sums1 = _mm_add_ps(m_sums1, m_sums2);
+
+ // Sum components together.
+ float result;
+ m_sums2 = _mm_add_ps(_mm_movehl_ps(m_sums1, m_sums1), m_sums1);
+ _mm_store_ss(&result, _mm_add_ss(m_sums2, _mm_shuffle_ps(
+ m_sums2, m_sums2, 1)));
+
+ return result;
+}
+
+} // namespace media
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/simd/vector_math_sse.cc b/src/3rdparty/chromium/media/base/simd/vector_math_sse.cc
new file mode 100644
index 000000000..c2121225c
--- /dev/null
+++ b/src/3rdparty/chromium/media/base/simd/vector_math_sse.cc
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
@@ -0,0 +1,118 @@
+// Copyright 2013 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "media/base/vector_math_testing.h"
+
+#include <algorithm>
+
+#include <xmmintrin.h> // NOLINT
+
+namespace media {
+namespace vector_math {
+
+void FMUL_SSE(const float src[], float scale, int len, float dest[]) {
+ const int rem = len % 4;
+ const int last_index = len - rem;
+ __m128 m_scale = _mm_set_ps1(scale);
+ for (int i = 0; i < last_index; i += 4)
+ _mm_store_ps(dest + i, _mm_mul_ps(_mm_load_ps(src + i), m_scale));
+
+ // Handle any remaining values that wouldn't fit in an SSE pass.
+ for (int i = last_index; i < len; ++i)
+ dest[i] = src[i] * scale;
+}
+
+void FMAC_SSE(const float src[], float scale, int len, float dest[]) {
+ const int rem = len % 4;
+ const int last_index = len - rem;
+ __m128 m_scale = _mm_set_ps1(scale);
+ for (int i = 0; i < last_index; i += 4) {
+ _mm_store_ps(dest + i, _mm_add_ps(_mm_load_ps(dest + i),
+ _mm_mul_ps(_mm_load_ps(src + i), m_scale)));
+ }
+
+ // Handle any remaining values that wouldn't fit in an SSE pass.
+ for (int i = last_index; i < len; ++i)
+ dest[i] += src[i] * scale;
+}
+
+// Convenience macro to extract float 0 through 3 from the vector |a|. This is
+// needed because compilers other than clang don't support access via
+// operator[]().
+#define EXTRACT_FLOAT(a, i) \
+ (i == 0 ? \
+ _mm_cvtss_f32(a) : \
+ _mm_cvtss_f32(_mm_shuffle_ps(a, a, i)))
+
+std::pair<float, float> EWMAAndMaxPower_SSE(
+ float initial_value, const float src[], int len, float smoothing_factor) {
+ // When the recurrence is unrolled, we see that we can split it into 4
+ // separate lanes of evaluation:
+ //
+ // y[n] = a(S[n]^2) + (1-a)(y[n-1])
+ // = a(S[n]^2) + (1-a)^1(aS[n-1]^2) + (1-a)^2(aS[n-2]^2) + ...
+ // = z[n] + (1-a)^1(z[n-1]) + (1-a)^2(z[n-2]) + (1-a)^3(z[n-3])
+ //
+ // where z[n] = a(S[n]^2) + (1-a)^4(z[n-4]) + (1-a)^8(z[n-8]) + ...
+ //
+ // Thus, the strategy here is to compute z[n], z[n-1], z[n-2], and z[n-3] in
+ // each of the 4 lanes, and then combine them to give y[n].
+
+ const int rem = len % 4;
+ const int last_index = len - rem;
+
+ const __m128 smoothing_factor_x4 = _mm_set_ps1(smoothing_factor);
+ const float weight_prev = 1.0f - smoothing_factor;
+ const __m128 weight_prev_x4 = _mm_set_ps1(weight_prev);
+ const __m128 weight_prev_squared_x4 =
+ _mm_mul_ps(weight_prev_x4, weight_prev_x4);
+ const __m128 weight_prev_4th_x4 =
+ _mm_mul_ps(weight_prev_squared_x4, weight_prev_squared_x4);
+
+ // Compute z[n], z[n-1], z[n-2], and z[n-3] in parallel in lanes 3, 2, 1 and
+ // 0, respectively.
+ __m128 max_x4 = _mm_setzero_ps();
+ __m128 ewma_x4 = _mm_setr_ps(0.0f, 0.0f, 0.0f, initial_value);
+ int i;
+ for (i = 0; i < last_index; i += 4) {
+ ewma_x4 = _mm_mul_ps(ewma_x4, weight_prev_4th_x4);
+ const __m128 sample_x4 = _mm_load_ps(src + i);
+ const __m128 sample_squared_x4 = _mm_mul_ps(sample_x4, sample_x4);
+ max_x4 = _mm_max_ps(max_x4, sample_squared_x4);
+ // Note: The compiler optimizes this to a single multiply-and-accumulate
+ // instruction:
+ ewma_x4 = _mm_add_ps(ewma_x4,
+ _mm_mul_ps(sample_squared_x4, smoothing_factor_x4));
+ }
+
+ // y[n] = z[n] + (1-a)^1(z[n-1]) + (1-a)^2(z[n-2]) + (1-a)^3(z[n-3])
+ float ewma = EXTRACT_FLOAT(ewma_x4, 3);
+ ewma_x4 = _mm_mul_ps(ewma_x4, weight_prev_x4);
+ ewma += EXTRACT_FLOAT(ewma_x4, 2);
+ ewma_x4 = _mm_mul_ps(ewma_x4, weight_prev_x4);
+ ewma += EXTRACT_FLOAT(ewma_x4, 1);
+ ewma_x4 = _mm_mul_ss(ewma_x4, weight_prev_x4);
+ ewma += EXTRACT_FLOAT(ewma_x4, 0);
+
+ // Fold the maximums together to get the overall maximum.
+ max_x4 = _mm_max_ps(max_x4,
+ _mm_shuffle_ps(max_x4, max_x4, _MM_SHUFFLE(3, 3, 1, 1)));
+ max_x4 = _mm_max_ss(max_x4, _mm_shuffle_ps(max_x4, max_x4, 2));
+
+ std::pair<float, float> result(ewma, EXTRACT_FLOAT(max_x4, 0));
+
+ // Handle remaining values at the end of |src|.
+ for (; i < len; ++i) {
+ result.first *= weight_prev;
+ const float sample = src[i];
+ const float sample_squared = sample * sample;
+ result.first += sample_squared * smoothing_factor;
+ result.second = std::max(result.second, sample_squared);
+ }
+
+ return result;
+}
+
+} // namespace vector_math
+} // namespace media
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/sinc_resampler.cc b/src/3rdparty/chromium/media/base/sinc_resampler.cc
index cffb0c9d6..010775544 100644
--- a/src/3rdparty/chromium/media/base/sinc_resampler.cc
+++ b/src/3rdparty/chromium/media/base/sinc_resampler.cc
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
@@ -81,17 +81,12 @@
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#include <cmath>
#include <limits>
+#include "base/cpu.h"
#include "base/logging.h"
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "build/build_config.h"
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
-#if defined(ARCH_CPU_X86_FAMILY)
-#include <xmmintrin.h>
-#define CONVOLVE_FUNC Convolve_SSE
-#elif defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
+#if defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
#include <arm_neon.h>
-#define CONVOLVE_FUNC Convolve_NEON
-#else
-#define CONVOLVE_FUNC Convolve_C
#endif
namespace media {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -112,10 +107,41 @@ static double SincScaleFactor(double io_ratio) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
return sinc_scale_factor;
}
+#undef CONVOLVE_FUNC
+
static int CalculateChunkSize(int block_size_, double io_ratio) {
return block_size_ / io_ratio;
}
+// If we know the minimum architecture at compile time, avoid CPU detection.
+// Force NaCl code to use C routines since (at present) nothing there uses these
+// methods and plumbing the -msse built library is non-trivial.
+#if defined(ARCH_CPU_X86_FAMILY) && !defined(OS_NACL)
+#if defined(__SSE__)
+#define CONVOLVE_FUNC Convolve_SSE
+void SincResampler::InitializeCPUSpecificFeatures() {}
+#else
+// X86 CPU detection required. Functions will be set by
+// InitializeCPUSpecificFeatures().
+#define CONVOLVE_FUNC g_convolve_proc_
+
+typedef float (*ConvolveProc)(const float*, const float*, const float*, double);
+static ConvolveProc g_convolve_proc_ = NULL;
+
+void SincResampler::InitializeCPUSpecificFeatures() {
+ CHECK(!g_convolve_proc_);
+ g_convolve_proc_ = base::CPU().has_sse() ? Convolve_SSE : Convolve_C;
+}
+#endif
+#elif defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
+#define CONVOLVE_FUNC Convolve_NEON
+void SincResampler::InitializeCPUSpecificFeatures() {}
+#else
+// Unknown architecture.
+#define CONVOLVE_FUNC Convolve_C
+void SincResampler::InitializeCPUSpecificFeatures() {}
+#endif
+
SincResampler::SincResampler(double io_sample_rate_ratio,
int request_frames,
const ReadCB& read_cb)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -328,46 +354,7 @@ float SincResampler::Convolve_C(const float* input_ptr, const float* k1,
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
kernel_interpolation_factor * sum2);
}
-#if defined(ARCH_CPU_X86_FAMILY)
-float SincResampler::Convolve_SSE(const float* input_ptr, const float* k1,
- const float* k2,
- double kernel_interpolation_factor) {
- __m128 m_input;
- __m128 m_sums1 = _mm_setzero_ps();
- __m128 m_sums2 = _mm_setzero_ps();
-
- // Based on |input_ptr| alignment, we need to use loadu or load. Unrolling
- // these loops hurt performance in local testing.
- if (reinterpret_cast<uintptr_t>(input_ptr) & 0x0F) {
- for (int i = 0; i < kKernelSize; i += 4) {
- m_input = _mm_loadu_ps(input_ptr + i);
- m_sums1 = _mm_add_ps(m_sums1, _mm_mul_ps(m_input, _mm_load_ps(k1 + i)));
- m_sums2 = _mm_add_ps(m_sums2, _mm_mul_ps(m_input, _mm_load_ps(k2 + i)));
- }
- } else {
- for (int i = 0; i < kKernelSize; i += 4) {
- m_input = _mm_load_ps(input_ptr + i);
- m_sums1 = _mm_add_ps(m_sums1, _mm_mul_ps(m_input, _mm_load_ps(k1 + i)));
- m_sums2 = _mm_add_ps(m_sums2, _mm_mul_ps(m_input, _mm_load_ps(k2 + i)));
- }
- }
-
- // Linearly interpolate the two "convolutions".
- m_sums1 = _mm_mul_ps(m_sums1, _mm_set_ps1(
- static_cast<float>(1.0 - kernel_interpolation_factor)));
- m_sums2 = _mm_mul_ps(m_sums2, _mm_set_ps1(
- static_cast<float>(kernel_interpolation_factor)));
- m_sums1 = _mm_add_ps(m_sums1, m_sums2);
-
- // Sum components together.
- float result;
- m_sums2 = _mm_add_ps(_mm_movehl_ps(m_sums1, m_sums1), m_sums1);
- _mm_store_ss(&result, _mm_add_ss(m_sums2, _mm_shuffle_ps(
- m_sums2, m_sums2, 1)));
-
- return result;
-}
-#elif defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
+#if defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
float SincResampler::Convolve_NEON(const float* input_ptr, const float* k1,
const float* k2,
double kernel_interpolation_factor) {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/sinc_resampler.h b/src/3rdparty/chromium/media/base/sinc_resampler.h
index afbd2abc9..9e3df2c55 100644
--- a/src/3rdparty/chromium/media/base/sinc_resampler.h
+++ b/src/3rdparty/chromium/media/base/sinc_resampler.h
@@ -36,6 +36,10 @@ class MEDIA_EXPORT SincResampler {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
kKernelStorageSize = kKernelSize * (kKernelOffsetCount + 1),
};
+ // Selects runtime specific CPU features like SSE. Must be called before
+ // using SincResampler.
+ static void InitializeCPUSpecificFeatures();
+
// Callback type for providing more data into the resampler. Expects |frames|
// of data to be rendered into |destination|; zero padded if not enough frames
// are available to satisfy the request.
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/sinc_resampler_perftest.cc b/src/3rdparty/chromium/media/base/sinc_resampler_perftest.cc
index 9cb7f4f2e..285382490 100644
--- a/src/3rdparty/chromium/media/base/sinc_resampler_perftest.cc
+++ b/src/3rdparty/chromium/media/base/sinc_resampler_perftest.cc
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
@@ -4,6 +4,7 @@
#include "base/bind.h"
#include "base/bind_helpers.h"
+#include "base/cpu.h"
#include "base/time/time.h"
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "build/build_config.h"
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#include "media/base/sinc_resampler.h"
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -61,6 +62,9 @@ TEST(SincResamplerPerfTest, Convolve) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
&resampler, SincResampler::Convolve_C, true, "unoptimized_aligned");
#if defined(CONVOLVE_FUNC)
+#if defined(ARCH_CPU_X86_FAMILY)
+ ASSERT_TRUE(base::CPU().has_sse());
+#endif
RunConvolveBenchmark(
&resampler, SincResampler::CONVOLVE_FUNC, true, "optimized_aligned");
RunConvolveBenchmark(
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/sinc_resampler_unittest.cc b/src/3rdparty/chromium/media/base/sinc_resampler_unittest.cc
index 8dd346ee4..7401eee96 100644
--- a/src/3rdparty/chromium/media/base/sinc_resampler_unittest.cc
+++ b/src/3rdparty/chromium/media/base/sinc_resampler_unittest.cc
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
@@ -10,6 +10,7 @@
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#include "base/bind.h"
#include "base/bind_helpers.h"
+#include "base/cpu.h"
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "base/macros.h"
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#include "base/strings/string_number_conversions.h"
#include "base/time/time.h"
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -166,6 +167,10 @@ TEST(SincResamplerTest, DISABLED_SetRatioBench) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
static const double kKernelInterpolationFactor = 0.5;
TEST(SincResamplerTest, Convolve) {
+#if defined(ARCH_CPU_X86_FAMILY)
+ ASSERT_TRUE(base::CPU().has_sse());
+#endif
+
// Initialize a dummy resampler.
MockSource mock_source;
SincResampler resampler(
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/vector_math.cc b/src/3rdparty/chromium/media/base/vector_math.cc
index 578290538..e041d8d4c 100644
--- a/src/3rdparty/chromium/media/base/vector_math.cc
+++ b/src/3rdparty/chromium/media/base/vector_math.cc
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
@@ -7,12 +7,17 @@
#include <algorithm>
+#include "base/cpu.h"
#include "base/logging.h"
#include "build/build_config.h"
+namespace media {
+namespace vector_math {
+
+// If we know the minimum architecture at compile time, avoid CPU detection.
// NaCl does not allow intrinsics.
#if defined(ARCH_CPU_X86_FAMILY) && !defined(OS_NACL)
-#include <xmmintrin.h>
+#if defined(__SSE__)
// Don't use custom SSE versions where the auto-vectorized C version performs
// better, which is anywhere clang is used.
#if !defined(__clang__)
@@ -23,20 +28,52 @@
#define FMUL_FUNC FMUL_C
#endif
#define EWMAAndMaxPower_FUNC EWMAAndMaxPower_SSE
+void Initialize() {}
+#else
+// X86 CPU detection required. Functions will be set by Initialize().
+#if !defined(__clang__)
+#define FMAC_FUNC g_fmac_proc_
+#define FMUL_FUNC g_fmul_proc_
+#else
+#define FMAC_FUNC FMAC_C
+#define FMUL_FUNC FMUL_C
+#endif
+#define EWMAAndMaxPower_FUNC g_ewma_power_proc_
+
+#if !defined(__clang__)
+typedef void (*MathProc)(const float src[], float scale, int len, float dest[]);
+static MathProc g_fmac_proc_ = NULL;
+static MathProc g_fmul_proc_ = NULL;
+#endif
+typedef std::pair<float, float> (*EWMAAndMaxPowerProc)(
+ float initial_value, const float src[], int len, float smoothing_factor);
+static EWMAAndMaxPowerProc g_ewma_power_proc_ = NULL;
+
+void Initialize() {
+ CHECK(!g_fmac_proc_);
+ CHECK(!g_fmul_proc_);
+ CHECK(!g_ewma_power_proc_);
+ const bool kUseSSE = base::CPU().has_sse();
+#if !defined(__clang__)
+ g_fmac_proc_ = kUseSSE ? FMAC_SSE : FMAC_C;
+ g_fmul_proc_ = kUseSSE ? FMUL_SSE : FMUL_C;
+#endif
+ g_ewma_power_proc_ = kUseSSE ? EWMAAndMaxPower_SSE : EWMAAndMaxPower_C;
+}
+#endif
#elif defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
#include <arm_neon.h>
#define FMAC_FUNC FMAC_NEON
#define FMUL_FUNC FMUL_NEON
#define EWMAAndMaxPower_FUNC EWMAAndMaxPower_NEON
+void Initialize() {}
#else
#define FMAC_FUNC FMAC_C
#define FMUL_FUNC FMUL_C
#define EWMAAndMaxPower_FUNC EWMAAndMaxPower_C
+void Initialize() {}
#endif
-namespace media {
-namespace vector_math {
-
void FMAC(const float src[], float scale, int len, float dest[]) {
// Ensure |src| and |dest| are 16-byte aligned.
DCHECK_EQ(0u, reinterpret_cast<uintptr_t>(src) & (kRequiredAlignment - 1));
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -89,111 +126,6 @@ std::pair<float, float> EWMAAndMaxPower_C(
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
return result;
}
-#if defined(ARCH_CPU_X86_FAMILY) && !defined(OS_NACL)
-void FMUL_SSE(const float src[], float scale, int len, float dest[]) {
- const int rem = len % 4;
- const int last_index = len - rem;
- __m128 m_scale = _mm_set_ps1(scale);
- for (int i = 0; i < last_index; i += 4)
- _mm_store_ps(dest + i, _mm_mul_ps(_mm_load_ps(src + i), m_scale));
-
- // Handle any remaining values that wouldn't fit in an SSE pass.
- for (int i = last_index; i < len; ++i)
- dest[i] = src[i] * scale;
-}
-
-void FMAC_SSE(const float src[], float scale, int len, float dest[]) {
- const int rem = len % 4;
- const int last_index = len - rem;
- __m128 m_scale = _mm_set_ps1(scale);
- for (int i = 0; i < last_index; i += 4) {
- _mm_store_ps(dest + i, _mm_add_ps(_mm_load_ps(dest + i),
- _mm_mul_ps(_mm_load_ps(src + i), m_scale)));
- }
-
- // Handle any remaining values that wouldn't fit in an SSE pass.
- for (int i = last_index; i < len; ++i)
- dest[i] += src[i] * scale;
-}
-
-// Convenience macro to extract float 0 through 3 from the vector |a|. This is
-// needed because compilers other than clang don't support access via
-// operator[]().
-#define EXTRACT_FLOAT(a, i) \
- (i == 0 ? \
- _mm_cvtss_f32(a) : \
- _mm_cvtss_f32(_mm_shuffle_ps(a, a, i)))
-
-std::pair<float, float> EWMAAndMaxPower_SSE(
- float initial_value, const float src[], int len, float smoothing_factor) {
- // When the recurrence is unrolled, we see that we can split it into 4
- // separate lanes of evaluation:
- //
- // y[n] = a(S[n]^2) + (1-a)(y[n-1])
- // = a(S[n]^2) + (1-a)^1(aS[n-1]^2) + (1-a)^2(aS[n-2]^2) + ...
- // = z[n] + (1-a)^1(z[n-1]) + (1-a)^2(z[n-2]) + (1-a)^3(z[n-3])
- //
- // where z[n] = a(S[n]^2) + (1-a)^4(z[n-4]) + (1-a)^8(z[n-8]) + ...
- //
- // Thus, the strategy here is to compute z[n], z[n-1], z[n-2], and z[n-3] in
- // each of the 4 lanes, and then combine them to give y[n].
-
- const int rem = len % 4;
- const int last_index = len - rem;
-
- const __m128 smoothing_factor_x4 = _mm_set_ps1(smoothing_factor);
- const float weight_prev = 1.0f - smoothing_factor;
- const __m128 weight_prev_x4 = _mm_set_ps1(weight_prev);
- const __m128 weight_prev_squared_x4 =
- _mm_mul_ps(weight_prev_x4, weight_prev_x4);
- const __m128 weight_prev_4th_x4 =
- _mm_mul_ps(weight_prev_squared_x4, weight_prev_squared_x4);
-
- // Compute z[n], z[n-1], z[n-2], and z[n-3] in parallel in lanes 3, 2, 1 and
- // 0, respectively.
- __m128 max_x4 = _mm_setzero_ps();
- __m128 ewma_x4 = _mm_setr_ps(0.0f, 0.0f, 0.0f, initial_value);
- int i;
- for (i = 0; i < last_index; i += 4) {
- ewma_x4 = _mm_mul_ps(ewma_x4, weight_prev_4th_x4);
- const __m128 sample_x4 = _mm_load_ps(src + i);
- const __m128 sample_squared_x4 = _mm_mul_ps(sample_x4, sample_x4);
- max_x4 = _mm_max_ps(max_x4, sample_squared_x4);
- // Note: The compiler optimizes this to a single multiply-and-accumulate
- // instruction:
- ewma_x4 = _mm_add_ps(ewma_x4,
- _mm_mul_ps(sample_squared_x4, smoothing_factor_x4));
- }
-
- // y[n] = z[n] + (1-a)^1(z[n-1]) + (1-a)^2(z[n-2]) + (1-a)^3(z[n-3])
- float ewma = EXTRACT_FLOAT(ewma_x4, 3);
- ewma_x4 = _mm_mul_ps(ewma_x4, weight_prev_x4);
- ewma += EXTRACT_FLOAT(ewma_x4, 2);
- ewma_x4 = _mm_mul_ps(ewma_x4, weight_prev_x4);
- ewma += EXTRACT_FLOAT(ewma_x4, 1);
- ewma_x4 = _mm_mul_ss(ewma_x4, weight_prev_x4);
- ewma += EXTRACT_FLOAT(ewma_x4, 0);
-
- // Fold the maximums together to get the overall maximum.
- max_x4 = _mm_max_ps(max_x4,
- _mm_shuffle_ps(max_x4, max_x4, _MM_SHUFFLE(3, 3, 1, 1)));
- max_x4 = _mm_max_ss(max_x4, _mm_shuffle_ps(max_x4, max_x4, 2));
-
- std::pair<float, float> result(ewma, EXTRACT_FLOAT(max_x4, 0));
-
- // Handle remaining values at the end of |src|.
- for (; i < len; ++i) {
- result.first *= weight_prev;
- const float sample = src[i];
- const float sample_squared = sample * sample;
- result.first += sample_squared * smoothing_factor;
- result.second = std::max(result.second, sample_squared);
- }
-
- return result;
-}
-#endif
-
#if defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
void FMAC_NEON(const float src[], float scale, int len, float dest[]) {
const int rem = len % 4;
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/vector_math.h b/src/3rdparty/chromium/media/base/vector_math.h
index a148ca050..0a2cb06f6 100644
--- a/src/3rdparty/chromium/media/base/vector_math.h
+++ b/src/3rdparty/chromium/media/base/vector_math.h
@@ -15,6 +15,11 @@ namespace vector_math {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
// Required alignment for inputs and outputs to all vector math functions
enum { kRequiredAlignment = 16 };
+// Selects runtime specific optimizations such as SSE. Must be called prior to
+// calling FMAC() or FMUL(). Called during media library initialization; most
+// users should never have to call this.
+MEDIA_EXPORT void Initialize();
+
// Multiply each element of |src| (up to |len|) by |scale| and add to |dest|.
// |src| and |dest| must be aligned by kRequiredAlignment.
MEDIA_EXPORT void FMAC(const float src[], float scale, int len, float dest[]);
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/vector_math_perftest.cc b/src/3rdparty/chromium/media/base/vector_math_perftest.cc
index 59b259dc2..c4c89a3b5 100644
--- a/src/3rdparty/chromium/media/base/vector_math_perftest.cc
+++ b/src/3rdparty/chromium/media/base/vector_math_perftest.cc
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
@@ -5,6 +5,7 @@
#include <memory>
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "base/macros.h"
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+#include "base/cpu.h"
#include "base/memory/aligned_memory.h"
#include "base/time/time.h"
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
#include "build/build_config.h"
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -82,15 +83,11 @@ class VectorMathPerfTest : public testing::Test {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
DISALLOW_COPY_AND_ASSIGN(VectorMathPerfTest);
};
-// Define platform dependent function names for SIMD optimized methods.
+// Define platform independent function name for FMAC* perf tests.
#if defined(ARCH_CPU_X86_FAMILY)
#define FMAC_FUNC FMAC_SSE
-#define FMUL_FUNC FMUL_SSE
-#define EWMAAndMaxPower_FUNC EWMAAndMaxPower_SSE
#elif defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
#define FMAC_FUNC FMAC_NEON
-#define FMUL_FUNC FMUL_NEON
-#define EWMAAndMaxPower_FUNC EWMAAndMaxPower_NEON
#endif
// Benchmark for each optimized vector_math::FMAC() method.
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -99,6 +96,9 @@ TEST_F(VectorMathPerfTest, FMAC) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
RunBenchmark(
vector_math::FMAC_C, true, "vector_math_fmac", "unoptimized");
#if defined(FMAC_FUNC)
+#if defined(ARCH_CPU_X86_FAMILY)
+ ASSERT_TRUE(base::CPU().has_sse());
+#endif
// Benchmark FMAC_FUNC() with unaligned size.
ASSERT_NE((kVectorSize - 1) % (vector_math::kRequiredAlignment /
sizeof(float)), 0U);
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -112,12 +112,24 @@ TEST_F(VectorMathPerfTest, FMAC) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#endif
}
+#undef FMAC_FUNC
+
+// Define platform independent function name for FMULBenchmark* tests.
+#if defined(ARCH_CPU_X86_FAMILY)
+#define FMUL_FUNC FMUL_SSE
+#elif defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
+#define FMUL_FUNC FMUL_NEON
+#endif
+
// Benchmark for each optimized vector_math::FMUL() method.
TEST_F(VectorMathPerfTest, FMUL) {
// Benchmark FMUL_C().
RunBenchmark(
vector_math::FMUL_C, true, "vector_math_fmul", "unoptimized");
#if defined(FMUL_FUNC)
+#if defined(ARCH_CPU_X86_FAMILY)
+ ASSERT_TRUE(base::CPU().has_sse());
+#endif
// Benchmark FMUL_FUNC() with unaligned size.
ASSERT_NE((kVectorSize - 1) % (vector_math::kRequiredAlignment /
sizeof(float)), 0U);
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -131,6 +143,14 @@ TEST_F(VectorMathPerfTest, FMUL) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#endif
}
+#undef FMUL_FUNC
+
+#if defined(ARCH_CPU_X86_FAMILY)
+#define EWMAAndMaxPower_FUNC EWMAAndMaxPower_SSE
+#elif defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
+#define EWMAAndMaxPower_FUNC EWMAAndMaxPower_NEON
+#endif
+
// Benchmark for each optimized vector_math::EWMAAndMaxPower() method.
TEST_F(VectorMathPerfTest, EWMAAndMaxPower) {
// Benchmark EWMAAndMaxPower_C().
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -139,6 +159,9 @@ TEST_F(VectorMathPerfTest, EWMAAndMaxPower) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
"vector_math_ewma_and_max_power",
"unoptimized");
#if defined(EWMAAndMaxPower_FUNC)
+#if defined(ARCH_CPU_X86_FAMILY)
+ ASSERT_TRUE(base::CPU().has_sse());
+#endif
// Benchmark EWMAAndMaxPower_FUNC() with unaligned size.
ASSERT_NE((kVectorSize - 1) % (vector_math::kRequiredAlignment /
sizeof(float)), 0U);
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -156,4 +179,6 @@ TEST_F(VectorMathPerfTest, EWMAAndMaxPower) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#endif
}
+#undef EWMAAndMaxPower_FUNC
+
} // namespace media
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/vector_math_testing.h b/src/3rdparty/chromium/media/base/vector_math_testing.h
index 9240fbf54..b0b304409 100644
--- a/src/3rdparty/chromium/media/base/vector_math_testing.h
+++ b/src/3rdparty/chromium/media/base/vector_math_testing.h
@@ -19,7 +19,7 @@ MEDIA_EXPORT void FMUL_C(const float src[], float scale, int len, float dest[]);
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
MEDIA_EXPORT std::pair<float, float> EWMAAndMaxPower_C(
float initial_value, const float src[], int len, float smoothing_factor);
-#if defined(ARCH_CPU_X86_FAMILY) && !defined(OS_NACL)
+#if defined(ARCH_CPU_X86_FAMILY)
MEDIA_EXPORT void FMAC_SSE(const float src[], float scale, int len,
float dest[]);
MEDIA_EXPORT void FMUL_SSE(const float src[], float scale, int len,
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/vector_math_unittest.cc b/src/3rdparty/chromium/media/base/vector_math_unittest.cc
index 3fcb3fad4..c4e8586aa 100644
--- a/src/3rdparty/chromium/media/base/vector_math_unittest.cc
+++ b/src/3rdparty/chromium/media/base/vector_math_unittest.cc
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
@@ -9,6 +9,7 @@
#include <memory>
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "base/macros.h"
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+#include "base/cpu.h"
#include "base/memory/aligned_memory.h"
#include "base/strings/string_number_conversions.h"
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
#include "base/strings/stringize_macros.h"
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -78,6 +79,7 @@ TEST_F(VectorMathTest, FMAC) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#if defined(ARCH_CPU_X86_FAMILY)
{
+ ASSERT_TRUE(base::CPU().has_sse());
SCOPED_TRACE("FMAC_SSE");
FillTestVectors(kInputFillValue, kOutputFillValue);
vector_math::FMAC_SSE(
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -119,6 +121,7 @@ TEST_F(VectorMathTest, FMUL) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#if defined(ARCH_CPU_X86_FAMILY)
{
+ ASSERT_TRUE(base::CPU().has_sse());
SCOPED_TRACE("FMUL_SSE");
FillTestVectors(kInputFillValue, kOutputFillValue);
vector_math::FMUL_SSE(
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -227,6 +230,7 @@ class EWMATestScenario {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#if defined(ARCH_CPU_X86_FAMILY)
{
+ ASSERT_TRUE(base::CPU().has_sse());
SCOPED_TRACE("EWMAAndMaxPower_SSE");
const std::pair<float, float>& result = vector_math::EWMAAndMaxPower_SSE(
initial_value_, data_.get(), data_len_, smoothing_factor_);
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/yuv_convert.cc b/src/3rdparty/chromium/media/base/yuv_convert.cc
index 2cdd9b1fd..2fc643416 100644
--- a/src/3rdparty/chromium/media/base/yuv_convert.cc
+++ b/src/3rdparty/chromium/media/base/yuv_convert.cc
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
@@ -32,7 +32,7 @@
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#include "media/base/simd/convert_yuv_to_rgb.h"
#include "media/base/simd/filter_yuv.h"
-#if defined(ARCH_CPU_X86_FAMILY)
+#if defined(ARCH_CPU_X86_FAMILY) && defined(__MMX__)
#if defined(COMPILER_MSVC)
#include <intrin.h>
#else
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -133,7 +133,7 @@ static const int16_t* g_table_rec709_ptr = NULL;
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
// Empty SIMD registers state after using them.
void EmptyRegisterStateStub() {}
-#if defined(MEDIA_MMX_INTRINSICS_AVAILABLE)
+#if defined(MEDIA_MMX_INTRINSICS_AVAILABLE) && defined(__MMX__)
void EmptyRegisterStateIntrinsic() { _mm_empty(); }
#endif
typedef void (*EmptyRegisterStateProc)();
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -247,34 +247,46 @@ void InitializeCPUSpecificYUVConversions() {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
// Assembly code confuses MemorySanitizer. Also not available in iOS builds.
#if defined(ARCH_CPU_X86_FAMILY) && !defined(MEMORY_SANITIZER) && \
!defined(OS_IOS)
- g_convert_yuva_to_argb_proc_ = ConvertYUVAToARGB_MMX;
+ base::CPU cpu;
+ if (cpu.has_mmx()) {
+ g_convert_yuv_to_rgb32_row_proc_ = ConvertYUVToRGB32Row_MMX;
+ g_scale_yuv_to_rgb32_row_proc_ = ScaleYUVToRGB32Row_MMX;
+ g_convert_yuv_to_rgb32_proc_ = ConvertYUVToRGB32_MMX;
+ g_convert_yuva_to_argb_proc_ = ConvertYUVAToARGB_MMX;
+ g_linear_scale_yuv_to_rgb32_row_proc_ = LinearScaleYUVToRGB32Row_MMX;
#if defined(MEDIA_MMX_INTRINSICS_AVAILABLE)
- g_empty_register_state_proc_ = EmptyRegisterStateIntrinsic;
+ g_filter_yuv_rows_proc_ = FilterYUVRows_MMX;
+#endif
+#if defined(MEDIA_MMX_INTRINSICS_AVAILABLE) && defined(__MMX__)
+ g_empty_register_state_proc_ = EmptyRegisterStateIntrinsic;
#else
- g_empty_register_state_proc_ = EmptyRegisterState_MMX;
+ g_empty_register_state_proc_ = EmptyRegisterState_MMX;
#endif
+ }
- g_convert_yuv_to_rgb32_row_proc_ = ConvertYUVToRGB32Row_SSE;
- g_convert_yuv_to_rgb32_proc_ = ConvertYUVToRGB32_SSE;
+ if (cpu.has_sse()) {
+ g_convert_yuv_to_rgb32_row_proc_ = ConvertYUVToRGB32Row_SSE;
+ g_scale_yuv_to_rgb32_row_proc_ = ScaleYUVToRGB32Row_SSE;
+ g_linear_scale_yuv_to_rgb32_row_proc_ = LinearScaleYUVToRGB32Row_SSE;
+ g_convert_yuv_to_rgb32_proc_ = ConvertYUVToRGB32_SSE;
+ }
- g_filter_yuv_rows_proc_ = FilterYUVRows_SSE2;
- g_convert_rgb32_to_yuv_proc_ = ConvertRGB32ToYUV_SSE2;
+ if (cpu.has_sse2()) {
+ g_filter_yuv_rows_proc_ = FilterYUVRows_SSE2;
+ g_convert_rgb32_to_yuv_proc_ = ConvertRGB32ToYUV_SSE2;
#if defined(ARCH_CPU_X86_64)
- g_scale_yuv_to_rgb32_row_proc_ = ScaleYUVToRGB32Row_SSE2_X64;
+ g_scale_yuv_to_rgb32_row_proc_ = ScaleYUVToRGB32Row_SSE2_X64;
- // Technically this should be in the MMX section, but MSVC will optimize out
- // the export of LinearScaleYUVToRGB32Row_MMX, which is required by the unit
- // tests, if that decision can be made at compile time. Since all X64 CPUs
- // have SSE2, we can hack around this by making the selection here.
- g_linear_scale_yuv_to_rgb32_row_proc_ = LinearScaleYUVToRGB32Row_MMX_X64;
-#else
- g_scale_yuv_to_rgb32_row_proc_ = ScaleYUVToRGB32Row_SSE;
- g_linear_scale_yuv_to_rgb32_row_proc_ = LinearScaleYUVToRGB32Row_SSE;
+ // Technically this should be in the MMX section, but MSVC will optimize out
+ // the export of LinearScaleYUVToRGB32Row_MMX, which is required by the unit
+ // tests, if that decision can be made at compile time. Since all X64 CPUs
+ // have SSE2, we can hack around this by making the selection here.
+ g_linear_scale_yuv_to_rgb32_row_proc_ = LinearScaleYUVToRGB32Row_MMX_X64;
#endif
+ }
- base::CPU cpu;
if (cpu.has_ssse3()) {
g_convert_rgb24_to_yuv_proc_ = &ConvertRGB24ToYUV_SSSE3;
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/yuv_convert_perftest.cc b/src/3rdparty/chromium/media/base/yuv_convert_perftest.cc
index ddf777cf3..f5520fd58 100644
--- a/src/3rdparty/chromium/media/base/yuv_convert_perftest.cc
+++ b/src/3rdparty/chromium/media/base/yuv_convert_perftest.cc
@@ -71,6 +71,29 @@ class YUVConvertPerfTest : public testing::Test {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
DISALLOW_COPY_AND_ASSIGN(YUVConvertPerfTest);
};
+TEST_F(YUVConvertPerfTest, ConvertYUVToRGB32Row_MMX) {
+ ASSERT_TRUE(base::CPU().has_mmx());
+
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ base::TimeTicks start = base::TimeTicks::Now();
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ for (int i = 0; i < kPerfTestIterations; ++i) {
+ for (int row = 0; row < kSourceHeight; ++row) {
+ int chroma_row = row / 2;
+ ConvertYUVToRGB32Row_MMX(
+ yuv_bytes_.get() + row * kSourceWidth,
+ yuv_bytes_.get() + kSourceUOffset + (chroma_row * kSourceWidth / 2),
+ yuv_bytes_.get() + kSourceVOffset + (chroma_row * kSourceWidth / 2),
+ rgb_bytes_converted_.get(),
+ kWidth,
+ GetLookupTable(YV12));
+ }
+ }
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ media::EmptyRegisterState();
+ double total_time_seconds = (base::TimeTicks::Now() - start).InSecondsF();
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ perf_test::PrintResult(
+ "yuv_convert_perftest", "", "ConvertYUVToRGB32Row_MMX",
+ kPerfTestIterations / total_time_seconds, "runs/s", true);
+}
+
TEST_F(YUVConvertPerfTest, ConvertYUVToRGB32Row_SSE) {
ASSERT_TRUE(base::CPU().has_sse());
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -161,9 +184,32 @@ TEST_F(YUVConvertPerfTest, I422AlphaToARGBRow_SSSE3) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
}
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#endif
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
-// 64-bit release + component builds on Windows are too smart and optimizes
-// away the function being tested.
-#if defined(OS_WIN) && (defined(ARCH_CPU_X86) || !defined(COMPONENT_BUILD))
+TEST_F(YUVConvertPerfTest, ScaleYUVToRGB32Row_MMX) {
+ ASSERT_TRUE(base::CPU().has_mmx());
+
+ const int kSourceDx = 80000; // This value means a scale down.
+
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ base::TimeTicks start = base::TimeTicks::Now();
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ for (int i = 0; i < kPerfTestIterations; ++i) {
+ for (int row = 0; row < kSourceHeight; ++row) {
+ int chroma_row = row / 2;
+ ScaleYUVToRGB32Row_MMX(
+ yuv_bytes_.get() + row * kSourceWidth,
+ yuv_bytes_.get() + kSourceUOffset + (chroma_row * kSourceWidth / 2),
+ yuv_bytes_.get() + kSourceVOffset + (chroma_row * kSourceWidth / 2),
+ rgb_bytes_converted_.get(),
+ kWidth,
+ kSourceDx,
+ GetLookupTable(YV12));
+ }
+ }
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ media::EmptyRegisterState();
+ double total_time_seconds = (base::TimeTicks::Now() - start).InSecondsF();
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ perf_test::PrintResult(
+ "yuv_convert_perftest", "", "ScaleYUVToRGB32Row_MMX",
+ kPerfTestIterations / total_time_seconds, "runs/s", true);
+}
+
TEST_F(YUVConvertPerfTest, ScaleYUVToRGB32Row_SSE) {
ASSERT_TRUE(base::CPU().has_sse());
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -190,6 +236,32 @@ TEST_F(YUVConvertPerfTest, ScaleYUVToRGB32Row_SSE) {
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
kPerfTestIterations / total_time_seconds, "runs/s", true);
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
}
+TEST_F(YUVConvertPerfTest, LinearScaleYUVToRGB32Row_MMX) {
+ ASSERT_TRUE(base::CPU().has_mmx());
+
+ const int kSourceDx = 80000; // This value means a scale down.
+
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ base::TimeTicks start = base::TimeTicks::Now();
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ for (int i = 0; i < kPerfTestIterations; ++i) {
+ for (int row = 0; row < kSourceHeight; ++row) {
+ int chroma_row = row / 2;
+ LinearScaleYUVToRGB32Row_MMX(
+ yuv_bytes_.get() + row * kSourceWidth,
+ yuv_bytes_.get() + kSourceUOffset + (chroma_row * kSourceWidth / 2),
+ yuv_bytes_.get() + kSourceVOffset + (chroma_row * kSourceWidth / 2),
+ rgb_bytes_converted_.get(),
+ kWidth,
+ kSourceDx,
+ GetLookupTable(YV12));
+ }
+ }
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ media::EmptyRegisterState();
+ double total_time_seconds = (base::TimeTicks::Now() - start).InSecondsF();
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ perf_test::PrintResult(
+ "yuv_convert_perftest", "", "LinearScaleYUVToRGB32Row_MMX",
+ kPerfTestIterations / total_time_seconds, "runs/s", true);
+}
+
TEST_F(YUVConvertPerfTest, LinearScaleYUVToRGB32Row_SSE) {
ASSERT_TRUE(base::CPU().has_sse());
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -215,7 +287,6 @@ TEST_F(YUVConvertPerfTest, LinearScaleYUVToRGB32Row_SSE) {
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
"yuv_convert_perftest", "", "LinearScaleYUVToRGB32Row_SSE",
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
kPerfTestIterations / total_time_seconds, "runs/s", true);
}
-#endif // defined(OS_WIN) && (ARCH_CPU_X86 || COMPONENT_BUILD)
#endif // !defined(ARCH_CPU_ARM_FAMILY) && !defined(ARCH_CPU_MIPS_FAMILY)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/media/base/yuv_convert_unittest.cc b/src/3rdparty/chromium/media/base/yuv_convert_unittest.cc
index ec872c89b..801466add 100644
--- a/src/3rdparty/chromium/media/base/yuv_convert_unittest.cc
+++ b/src/3rdparty/chromium/media/base/yuv_convert_unittest.cc
@@ -643,6 +643,37 @@ TEST(YUVConvertTest, RGB32ToYUV_SSE2_MatchReference) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
EXPECT_EQ(0, error);
}
+TEST(YUVConvertTest, ConvertYUVToRGB32Row_MMX) {
+ base::CPU cpu;
+ if (!cpu.has_mmx()) {
+ LOG(WARNING) << "System not supported. Test skipped.";
+ return;
+ }
+
+ scoped_ptr<uint8[]> yuv_bytes(new uint8[kYUV12Size]);
+ scoped_ptr<uint8[]> rgb_bytes_reference(new uint8[kRGBSize]);
+ scoped_ptr<uint8[]> rgb_bytes_converted(new uint8[kRGBSize]);
+ ReadYV12Data(&yuv_bytes);
+
+ const int kWidth = 167;
+ ConvertYUVToRGB32Row_C(yuv_bytes.get(),
+ yuv_bytes.get() + kSourceUOffset,
+ yuv_bytes.get() + kSourceVOffset,
+ rgb_bytes_reference.get(),
+ kWidth,
+ GetLookupTable(YV12));
+ ConvertYUVToRGB32Row_MMX(yuv_bytes.get(),
+ yuv_bytes.get() + kSourceUOffset,
+ yuv_bytes.get() + kSourceVOffset,
+ rgb_bytes_converted.get(),
+ kWidth,
+ GetLookupTable(YV12));
+ media::EmptyRegisterState();
+ EXPECT_EQ(0, memcmp(rgb_bytes_reference.get(),
+ rgb_bytes_converted.get(),
+ kWidth * kBpp));
+}
+
TEST(YUVConvertTest, ConvertYUVToRGB32Row_SSE) {
base::CPU cpu;
if (!cpu.has_sse()) {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -674,9 +705,40 @@ TEST(YUVConvertTest, ConvertYUVToRGB32Row_SSE) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
kWidth * kBpp));
}
-// 64-bit release + component builds on Windows are too smart and optimizes
-// away the function being tested.
-#if defined(OS_WIN) && (defined(ARCH_CPU_X86) || !defined(COMPONENT_BUILD))
+TEST(YUVConvertTest, ScaleYUVToRGB32Row_MMX) {
+ base::CPU cpu;
+ if (!cpu.has_mmx()) {
+ LOG(WARNING) << "System not supported. Test skipped.";
+ return;
+ }
+
+ scoped_ptr<uint8[]> yuv_bytes(new uint8[kYUV12Size]);
+ scoped_ptr<uint8[]> rgb_bytes_reference(new uint8[kRGBSize]);
+ scoped_ptr<uint8[]> rgb_bytes_converted(new uint8[kRGBSize]);
+ ReadYV12Data(&yuv_bytes);
+
+ const int kWidth = 167;
+ const int kSourceDx = 80000; // This value means a scale down.
+ ScaleYUVToRGB32Row_C(yuv_bytes.get(),
+ yuv_bytes.get() + kSourceUOffset,
+ yuv_bytes.get() + kSourceVOffset,
+ rgb_bytes_reference.get(),
+ kWidth,
+ kSourceDx,
+ GetLookupTable(YV12));
+ ScaleYUVToRGB32Row_MMX(yuv_bytes.get(),
+ yuv_bytes.get() + kSourceUOffset,
+ yuv_bytes.get() + kSourceVOffset,
+ rgb_bytes_converted.get(),
+ kWidth,
+ kSourceDx,
+ GetLookupTable(YV12));
+ media::EmptyRegisterState();
+ EXPECT_EQ(0, memcmp(rgb_bytes_reference.get(),
+ rgb_bytes_converted.get(),
+ kWidth * kBpp));
+}
+
TEST(YUVConvertTest, ScaleYUVToRGB32Row_SSE) {
base::CPU cpu;
if (!cpu.has_sse()) {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -711,6 +773,40 @@ TEST(YUVConvertTest, ScaleYUVToRGB32Row_SSE) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
kWidth * kBpp));
}
+TEST(YUVConvertTest, LinearScaleYUVToRGB32Row_MMX) {
+ base::CPU cpu;
+ if (!cpu.has_mmx()) {
+ LOG(WARNING) << "System not supported. Test skipped.";
+ return;
+ }
+
+ scoped_ptr<uint8[]> yuv_bytes(new uint8[kYUV12Size]);
+ scoped_ptr<uint8[]> rgb_bytes_reference(new uint8[kRGBSize]);
+ scoped_ptr<uint8[]> rgb_bytes_converted(new uint8[kRGBSize]);
+ ReadYV12Data(&yuv_bytes);
+
+ const int kWidth = 167;
+ const int kSourceDx = 80000; // This value means a scale down.
+ LinearScaleYUVToRGB32Row_C(yuv_bytes.get(),
+ yuv_bytes.get() + kSourceUOffset,
+ yuv_bytes.get() + kSourceVOffset,
+ rgb_bytes_reference.get(),
+ kWidth,
+ kSourceDx,
+ GetLookupTable(YV12));
+ LinearScaleYUVToRGB32Row_MMX(yuv_bytes.get(),
+ yuv_bytes.get() + kSourceUOffset,
+ yuv_bytes.get() + kSourceVOffset,
+ rgb_bytes_converted.get(),
+ kWidth,
+ kSourceDx,
+ GetLookupTable(YV12));
+ media::EmptyRegisterState();
+ EXPECT_EQ(0, memcmp(rgb_bytes_reference.get(),
+ rgb_bytes_converted.get(),
+ kWidth * kBpp));
+}
+
TEST(YUVConvertTest, LinearScaleYUVToRGB32Row_SSE) {
base::CPU cpu;
if (!cpu.has_sse()) {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -744,7 +840,6 @@ TEST(YUVConvertTest, LinearScaleYUVToRGB32Row_SSE) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
rgb_bytes_converted.get(),
kWidth * kBpp));
}
-#endif // defined(OS_WIN) && (ARCH_CPU_X86 || COMPONENT_BUILD)
TEST(YUVConvertTest, FilterYUVRows_C_OutOfBounds) {
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
std::unique_ptr<uint8_t[]> src(new uint8_t[16]);
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -761,6 +856,30 @@ TEST(YUVConvertTest, FilterYUVRows_C_OutOfBounds) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
}
}
+#if defined(MEDIA_MMX_INTRINSICS_AVAILABLE)
+TEST(YUVConvertTest, FilterYUVRows_MMX_OutOfBounds) {
+ base::CPU cpu;
+ if (!cpu.has_mmx()) {
+ LOG(WARNING) << "System not supported. Test skipped.";
+ return;
+ }
+
+ scoped_ptr<uint8[]> src(new uint8[16]);
+ scoped_ptr<uint8[]> dst(new uint8[16]);
+
+ memset(src.get(), 0xff, 16);
+ memset(dst.get(), 0, 16);
+
+ media::FilterYUVRows_MMX(dst.get(), src.get(), src.get(), 1, 255);
+ media::EmptyRegisterState();
+
+ EXPECT_EQ(255u, dst[0]);
+ for (int i = 1; i < 16; ++i) {
+ EXPECT_EQ(0u, dst[i]);
+ }
+}
+#endif // defined(MEDIA_MMX_INTRINSICS_AVAILABLE)
+
TEST(YUVConvertTest, FilterYUVRows_SSE2_OutOfBounds) {
base::CPU cpu;
if (!cpu.has_sse2()) {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -782,6 +901,38 @@ TEST(YUVConvertTest, FilterYUVRows_SSE2_OutOfBounds) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
}
}
+#if defined(MEDIA_MMX_INTRINSICS_AVAILABLE)
+TEST(YUVConvertTest, FilterYUVRows_MMX_UnalignedDestination) {
+ base::CPU cpu;
+ if (!cpu.has_mmx()) {
+ LOG(WARNING) << "System not supported. Test skipped.";
+ return;
+ }
+
+ const int kSize = 32;
+ scoped_ptr<uint8[]> src(new uint8[kSize]);
+ scoped_ptr<uint8[]> dst_sample(new uint8[kSize]);
+ scoped_ptr<uint8[]> dst(new uint8[kSize]);
+
+ memset(dst_sample.get(), 0, kSize);
+ memset(dst.get(), 0, kSize);
+ for (int i = 0; i < kSize; ++i)
+ src[i] = 100 + i;
+
+ media::FilterYUVRows_C(dst_sample.get(),
+ src.get(), src.get(), 17, 128);
+
+ // Generate an unaligned output address.
+ uint8* dst_ptr =
+ reinterpret_cast<uint8*>(
+ (reinterpret_cast<uintptr_t>(dst.get() + 8) & ~7) + 1);
+ media::FilterYUVRows_MMX(dst_ptr, src.get(), src.get(), 17, 128);
+ media::EmptyRegisterState();
+
+ EXPECT_EQ(0, memcmp(dst_sample.get(), dst_ptr, 17));
+}
+#endif // defined(MEDIA_MMX_INTRINSICS_AVAILABLE)
+
TEST(YUVConvertTest, FilterYUVRows_SSE2_UnalignedDestination) {
base::CPU cpu;
if (!cpu.has_sse2()) {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/skia/BUILD.gn b/src/3rdparty/chromium/skia/BUILD.gn
index 55854ffe6..9a1d4ec0f 100644
--- a/src/3rdparty/chromium/skia/BUILD.gn
+++ b/src/3rdparty/chromium/skia/BUILD.gn
@@ -1,38 +1,3 @@
-# Copyright (c) 2013 The Chromium Authors. All rights reserved.
-# Use of this source code is governed by a BSD-style license that can be
-# found in the LICENSE file.
-
-import("//build/config/features.gni")
-import("//build/config/ui.gni")
-import("//printing/features/features.gni")
-import("//testing/test.gni")
-import("//third_party/skia/gn/shared_sources.gni")
-
-if (current_cpu == "arm") {
- import("//build/config/arm.gni")
-}
-if (current_cpu == "mipsel" || current_cpu == "mips64el") {
- import("//build/config/mips.gni")
-}
-
-skia_support_gpu = !is_ios
-skia_support_pdf = !is_ios && (enable_basic_printing || enable_print_preview)
-
-# When building Chrome for iOS with GYP, the target cpu is not known during
-# the invocation of gyp. This cause the iOS build to use non-optimised skia.
-# Replicate this with GN to avoid introducing regression as recommended by
-# the OWNERS of skia.
-
-declare_args() {
- skia_whitelist_serialized_typefaces = false
-
- # TODO(crbug.com/607933): Once GYP is no longer supported, port iOS to use
- # optimised skia.
- skia_build_no_opts = is_ios
-}
-
-# External-facing config for dependent code.
-config("skia_config") {
include_dirs = [
"config",
"ext",
@@ -608,7 +573,15 @@ source_set("skia_opts") {
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
if (skia_build_no_opts) {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
sources = skia_opts.none_sources
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
} else if (current_cpu == "x86" || current_cpu == "x64") {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
- sources = skia_opts.sse2_sources
+ sources = skia_opts.sse2_sources +
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ [
+ # Chrome-specific.
+ "ext/convolver_SSE2.cc",
+ "ext/convolver_SSE2.h",
+ ]
+ if (!is_win || is_clang) {
+ cflags += [ "-msse2" ]
+ }
deps += [
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
":skia_opts_avx",
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
":skia_opts_hsw",
@@ -644,6 +617,13 @@ source_set("skia_opts") {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
if (mips_dsp_rev >= 1) {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
sources = skia_opts.mips_dsp_sources
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+ if (mips_dsp_rev >= 2) {
+ sources += [
+ # Chrome-specific.
+ "ext/convolver_mips_dspr2.cc",
+ "ext/convolver_mips_dspr2.h",
+ ]
+ }
} else {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
sources = skia_opts.none_sources
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
}
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/skia/ext/convolver.cc b/src/3rdparty/chromium/skia/ext/convolver.cc
index 092fefaa9..4b40ffd2c 100644
--- a/src/3rdparty/chromium/skia/ext/convolver.cc
+++ b/src/3rdparty/chromium/skia/ext/convolver.cc
@@ -362,10 +362,13 @@ struct ConvolveProcs {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
void SetupSIMD(ConvolveProcs *procs) {
#ifdef SIMD_SSE2
- procs->extra_horizontal_reads = 3;
- procs->convolve_vertically = &ConvolveVertically_SSE2;
- procs->convolve_4rows_horizontally = &Convolve4RowsHorizontally_SSE2;
- procs->convolve_horizontally = &ConvolveHorizontally_SSE2;
+ base::CPU cpu;
+ if (cpu.has_sse2()) {
+ procs->extra_horizontal_reads = 3;
+ procs->convolve_vertically = &ConvolveVertically_SSE2;
+ procs->convolve_4rows_horizontally = &Convolve4RowsHorizontally_SSE2;
+ procs->convolve_horizontally = &ConvolveHorizontally_SSE2;
+ }
#elif defined SIMD_MIPS_DSPR2
procs->extra_horizontal_reads = 3;
procs->convolve_vertically = &ConvolveVertically_mips_dspr2;
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/skia/ext/convolver.h b/src/3rdparty/chromium/skia/ext/convolver.h
index 1f33d4c0f..f7cd4770a 100644
--- a/src/3rdparty/chromium/skia/ext/convolver.h
+++ b/src/3rdparty/chromium/skia/ext/convolver.h
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
@@ -11,6 +11,7 @@
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#include <vector>
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "build/build_config.h"
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
+#include "base/cpu.h"
#include "third_party/skia/include/core/SkSize.h"
#include "third_party/skia/include/core/SkTypes.h"
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/third_party/WebKit/Source/modules/webaudio/AudioParamTimeline.cpp b/src/3rdparty/chromium/third_party/WebKit/Source/modules/webaudio/AudioParamTimeline.cpp
index 8e558f439..20ba17688 100644
--- a/src/3rdparty/chromium/third_party/WebKit/Source/modules/webaudio/AudioParamTimeline.cpp
+++ b/src/3rdparty/chromium/third_party/WebKit/Source/modules/webaudio/AudioParamTimeline.cpp
@@ -31,7 +31,7 @@
no-sse2 patch: Also fix AudioParamTimeline.cpp in Blink's webaudio src/3rdparty/chromium/third_party/WebKit/Source/modules/webaudio/AudioParamTimeline.cpp
2016-07-18 23:24:13 +00:00
#include "wtf/MathExtras.h"
#include <algorithm>
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
#include <emmintrin.h>
#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -662,7 +662,7 @@ float AudioParamTimeline::valuesForFrameRangeImpl(size_t startFrame,
// the next event.
if (nextEventType == ParamEvent::LinearRampToValue) {
const float valueDelta = value2 - value1;
no-sse2 patch: Also fix AudioParamTimeline.cpp in Blink's webaudio src/3rdparty/chromium/third_party/WebKit/Source/modules/webaudio/AudioParamTimeline.cpp
2016-07-18 23:24:13 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
if (fillToFrame > writeIndex) {
// Minimize in-loop operations. Calculate starting value and increment.
// Next step: value += inc.
@@ -841,7 +841,7 @@ float AudioParamTimeline::valuesForFrameRangeImpl(size_t startFrame,
for (; writeIndex < fillToFrame; ++writeIndex)
values[writeIndex] = target;
} else {
no-sse2 patch: Also fix AudioParamTimeline.cpp in Blink's webaudio src/3rdparty/chromium/third_party/WebKit/Source/modules/webaudio/AudioParamTimeline.cpp
2016-07-18 23:24:13 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
if (fillToFrame > writeIndex) {
// Resolve recursion by expanding constants to achieve a 4-step
// loop unrolling.
@@ -959,7 +959,7 @@ float AudioParamTimeline::valuesForFrameRangeImpl(size_t startFrame,
// Oversampled curve data can be provided if sharp discontinuities are
// desired.
unsigned k = 0;
no-sse2 patch: Also fix AudioParamTimeline.cpp in Blink's webaudio src/3rdparty/chromium/third_party/WebKit/Source/modules/webaudio/AudioParamTimeline.cpp
2016-07-18 23:24:13 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
if (fillToFrame > writeIndex) {
const __m128 vCurveVirtualIndex = _mm_set_ps1(curveVirtualIndex);
const __m128 vCurvePointsPerFrame =
diff --git a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/DirectConvolver.cpp b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/DirectConvolver.cpp
index a77ec195c..d76c2d015 100644
--- a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/DirectConvolver.cpp
+++ b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/DirectConvolver.cpp
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
@@ -26,6 +26,9 @@
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+// include this first to get it before the CPU() function-like macro
+#include "base/cpu.h"
+
#include "platform/audio/DirectConvolver.h"
#if OS(MACOSX)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -35,21 +38,45 @@
no-sse2.patch: remove -ffloat-store, fix DirectConvolver FTBFS * Remove the -ffloat-store workaround: The media player issue that caused problems with x87 floating point (and ultimately made upstream require SSE2) should already be fixed / worked around by: https://crrev.com/d2c745b13c93ecff5108bed57d8e052126715492 So I will only readd -ffloat-store if people request it, and even then, only to specific parts of code and not everything. * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ DirectConvolver.cpp to not require SSE2 on x86. (No runtime detection, sorry.) * Update the changelog date in the specfile.
2016-01-17 01:34:08 +00:00
#include "platform/audio/VectorMath.h"
#include "wtf/CPU.h"
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
-#if (CPU(X86) || CPU(X86_64)) && !OS(MACOSX)
+#if ((CPU(X86) && defined(__SSE2__)) || CPU(X86_64)) && !OS(MACOSX)
no-sse2.patch: remove -ffloat-store, fix DirectConvolver FTBFS * Remove the -ffloat-store workaround: The media player issue that caused problems with x87 floating point (and ultimately made upstream require SSE2) should already be fixed / worked around by: https://crrev.com/d2c745b13c93ecff5108bed57d8e052126715492 So I will only readd -ffloat-store if people request it, and even then, only to specific parts of code and not everything. * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ DirectConvolver.cpp to not require SSE2 on x86. (No runtime detection, sorry.) * Update the changelog date in the specfile.
2016-01-17 01:34:08 +00:00
#include <emmintrin.h>
#endif
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if defined(BUILD_ONLY_THE_SSE2_PARTS) && !defined(__SSE2__)
+#error SSE2 parts must be built with -msse2
+#endif
+
namespace blink {
using namespace VectorMath;
+#ifndef BUILD_ONLY_THE_SSE2_PARTS
+
DirectConvolver::DirectConvolver(size_t inputBlockSize)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
- : m_inputBlockSize(inputBlockSize), m_buffer(inputBlockSize * 2) {}
+ : m_inputBlockSize(inputBlockSize), m_buffer(inputBlockSize * 2)
+{
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if CPU(X86)
+ base::CPU cpu;
+ m_haveSSE2 = cpu.has_sse2();
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+}
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#ifdef BUILD_ONLY_THE_SSE2_PARTS
+void DirectConvolver::m_processSSE2(AudioFloatArray* convolutionKernel, const float* sourceP, float* destP, size_t framesToProcess)
+#else
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
void DirectConvolver::process(AudioFloatArray* convolutionKernel,
const float* sourceP,
float* destP,
- size_t framesToProcess) {
+ size_t framesToProcess)
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+{
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+ if (m_haveSSE2) {
+ m_processSSE2(convolutionKernel, sourceP, destP, framesToProcess);
+ return;
+ }
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
ASSERT(framesToProcess == m_inputBlockSize);
if (framesToProcess != m_inputBlockSize)
return;
@@ -83,7 +110,7 @@ void DirectConvolver::process(AudioFloatArray* convolutionKernel,
#endif // CPU(X86)
no-sse2.patch: remove -ffloat-store, fix DirectConvolver FTBFS * Remove the -ffloat-store workaround: The media player issue that caused problems with x87 floating point (and ultimately made upstream require SSE2) should already be fixed / worked around by: https://crrev.com/d2c745b13c93ecff5108bed57d8e052126715492 So I will only readd -ffloat-store if people request it, and even then, only to specific parts of code and not everything. * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ DirectConvolver.cpp to not require SSE2 on x86. (No runtime detection, sorry.) * Update the changelog date in the specfile.
2016-01-17 01:34:08 +00:00
#else
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
size_t i = 0;
no-sse2.patch: remove -ffloat-store, fix DirectConvolver FTBFS * Remove the -ffloat-store workaround: The media player issue that caused problems with x87 floating point (and ultimately made upstream require SSE2) should already be fixed / worked around by: https://crrev.com/d2c745b13c93ecff5108bed57d8e052126715492 So I will only readd -ffloat-store if people request it, and even then, only to specific parts of code and not everything. * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ DirectConvolver.cpp to not require SSE2 on x86. (No runtime detection, sorry.) * Update the changelog date in the specfile.
2016-01-17 01:34:08 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
// Convolution using SSE2. Currently only do this if both |kernelSize| and
// |framesToProcess| are multiples of 4. If not, use the straightforward loop
// below.
@@ -397,7 +424,7 @@ void DirectConvolver::process(AudioFloatArray* convolutionKernel,
}
destP[i++] = sum;
no-sse2.patch: remove -ffloat-store, fix DirectConvolver FTBFS * Remove the -ffloat-store workaround: The media player issue that caused problems with x87 floating point (and ultimately made upstream require SSE2) should already be fixed / worked around by: https://crrev.com/d2c745b13c93ecff5108bed57d8e052126715492 So I will only readd -ffloat-store if people request it, and even then, only to specific parts of code and not everything. * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ DirectConvolver.cpp to not require SSE2 on x86. (No runtime detection, sorry.) * Update the changelog date in the specfile.
2016-01-17 01:34:08 +00:00
}
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
}
no-sse2.patch: remove -ffloat-store, fix DirectConvolver FTBFS * Remove the -ffloat-store workaround: The media player issue that caused problems with x87 floating point (and ultimately made upstream require SSE2) should already be fixed / worked around by: https://crrev.com/d2c745b13c93ecff5108bed57d8e052126715492 So I will only readd -ffloat-store if people request it, and even then, only to specific parts of code and not everything. * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ DirectConvolver.cpp to not require SSE2 on x86. (No runtime detection, sorry.) * Update the changelog date in the specfile.
2016-01-17 01:34:08 +00:00
#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
#endif // OS(MACOSX)
@@ -406,8 +433,9 @@ void DirectConvolver::process(AudioFloatArray* convolutionKernel,
memcpy(m_buffer.data(), inputP, sizeof(float) * framesToProcess);
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
}
+#ifndef BUILD_ONLY_THE_SSE2_PARTS
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
void DirectConvolver::reset() {
m_buffer.zero();
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
}
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
-
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
} // namespace blink
diff --git a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/DirectConvolver.h b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/DirectConvolver.h
index 848f6aa73..78ea0d84c 100644
--- a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/DirectConvolver.h
+++ b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/DirectConvolver.h
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
@@ -32,6 +32,7 @@
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
#include "platform/PlatformExport.h"
#include "platform/audio/AudioArray.h"
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "wtf/Allocator.h"
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#include "wtf/CPU.h"
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "wtf/Noncopyable.h"
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
namespace blink {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -54,6 +55,12 @@ class PLATFORM_EXPORT DirectConvolver {
size_t m_inputBlockSize;
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
AudioFloatArray m_buffer;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+
no-sse2.patch: Fix typos in the patched header files
2016-01-17 04:10:15 +00:00
+#if CPU(X86)
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+ bool m_haveSSE2;
+ void m_processSSE2(AudioFloatArray* convolutionKernel, const float* sourceP, float* destP, size_t framesToProcess);
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
};
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
} // namespace blink
diff --git a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/SincResampler.cpp b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/SincResampler.cpp
index e7ca2c77d..a0ad93953 100644
--- a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/SincResampler.cpp
+++ b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/SincResampler.cpp
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
@@ -26,15 +26,22 @@
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+// include this first to get it before the CPU() function-like macro
+#include "base/cpu.h"
+
#include "platform/audio/SincResampler.h"
#include "platform/audio/AudioBus.h"
#include "wtf/CPU.h"
#include "wtf/MathExtras.h"
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
#include <emmintrin.h>
#endif
+#if defined(BUILD_ONLY_THE_SSE2_PARTS) && !defined(__SSE2__)
+#error SSE2 parts must be built with -msse2
+#endif
+
// Input buffer layout, dividing the total buffer into regions (r0 - r5):
//
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
// |----------------|-----------------------------------------|----------------|
@@ -66,6 +73,8 @@
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
namespace blink {
+#ifndef BUILD_ONLY_THE_SSE2_PARTS
+
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
SincResampler::SincResampler(double scaleFactor,
unsigned kernelSize,
unsigned numberOfKernelOffsets)
@@ -81,6 +90,10 @@ SincResampler::SincResampler(double scaleFactor,
m_sourceFramesAvailable(0),
m_sourceProvider(nullptr),
m_isBufferPrimed(false) {
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if CPU(X86)
+ base::CPU cpu;
+ m_haveSSE2 = cpu.has_sse2();
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
initializeKernel();
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
}
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -200,10 +213,22 @@ void SincResampler::process(const float* source,
remaining -= framesThisTime;
}
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
}
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+#endif //BUILD_ONLY_THE_SSE2_PARTS
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#ifdef BUILD_ONLY_THE_SSE2_PARTS
+void SincResampler::m_processSSE2(AudioSourceProvider* sourceProvider, float* destination, size_t framesToProcess)
+#else
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
void SincResampler::process(AudioSourceProvider* sourceProvider,
float* destination,
- size_t framesToProcess) {
+ size_t framesToProcess)
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+{
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if CPU(X86) && !defined(__SSE2__)
+ if (m_haveSSE2) {
+ m_processSSE2(sourceProvider, destination, framesToProcess);
+ return;
+ }
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
bool isGood = sourceProvider && m_blockSize > m_kernelSize &&
m_inputBuffer.size() >= m_blockSize + m_kernelSize &&
!(m_kernelSize % 2);
@@ -269,7 +294,7 @@ void SincResampler::process(AudioSourceProvider* sourceProvider,
{
float input;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
// If the sourceP address is not 16-byte aligned, the first several
// frames (at most three) should be processed seperately.
while ((reinterpret_cast<uintptr_t>(inputP) & 0x0F) && n) {
diff --git a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/SincResampler.h b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/SincResampler.h
index b96ec94cf..4b6b3293e 100644
--- a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/SincResampler.h
+++ b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/SincResampler.h
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
@@ -33,6 +33,7 @@
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
#include "platform/audio/AudioArray.h"
#include "platform/audio/AudioSourceProvider.h"
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "wtf/Allocator.h"
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#include "wtf/CPU.h"
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "wtf/Noncopyable.h"
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
namespace blink {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -96,6 +97,12 @@ class PLATFORM_EXPORT SincResampler {
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
// The buffer is primed once at the very beginning of processing.
bool m_isBufferPrimed;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+
no-sse2.patch: Fix typos in the patched header files
2016-01-17 04:10:15 +00:00
+#if CPU(X86)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+ bool m_haveSSE2;
+ void m_processSSE2(AudioSourceProvider*, float* destination, size_t framesToProcess);
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+ };
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
};
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
} // namespace blink
diff --git a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/VectorMath.cpp b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/VectorMath.cpp
index 8f3592cd7..ba9fc1a73 100644
--- a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/VectorMath.cpp
+++ b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/VectorMath.cpp
@@ -23,6 +23,9 @@
* DAMAGE.
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
*/
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+// include this first to get it before the CPU() function-like macro
+#include "base/cpu.h"
+
#include "platform/audio/VectorMath.h"
#include "wtf/Assertions.h"
#include "wtf/CPU.h"
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -33,10 +36,14 @@
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
#include <Accelerate/Accelerate.h>
#endif
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
#include <emmintrin.h>
#endif
+#if defined(BUILD_ONLY_THE_SSE2_PARTS) && !defined(__SSE2__)
+#error SSE2 parts must be built with -msse2
+#endif
+
#if HAVE(ARM_NEON_INTRINSICS)
#include <arm_neon.h>
#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -165,15 +172,30 @@ void vclip(const float* sourceP,
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
}
#else
+#ifdef BUILD_ONLY_THE_SSE2_PARTS
+namespace SSE2 {
+#endif
+
+#if CPU(X86) && !defined(__SSE2__)
+static base::CPU cpu;
+#endif
+
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
void vsma(const float* sourceP,
int sourceStride,
const float* scale,
float* destP,
int destStride,
size_t framesToProcess) {
- int n = framesToProcess;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if CPU(X86) && !defined(__SSE2__)
+ if (cpu.has_sse2()) {
+ blink::VectorMath::SSE2::vsma(sourceP, sourceStride, scale, destP, destStride, framesToProcess);
+ return;
+ }
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+
+int n = framesToProcess;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
if ((sourceStride == 1) && (destStride == 1)) {
float k = *scale;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -269,9 +291,16 @@ void vsmul(const float* sourceP,
float* destP,
int destStride,
size_t framesToProcess) {
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if CPU(X86) && !defined(__SSE2__)
+ if (cpu.has_sse2()) {
+ blink::VectorMath::SSE2::vsmul(sourceP, sourceStride, scale, destP, destStride, framesToProcess);
+ return;
+ }
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+
int n = framesToProcess;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
if ((sourceStride == 1) && (destStride == 1)) {
float k = *scale;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -360,7 +389,7 @@ void vsmul(const float* sourceP,
sourceP += sourceStride;
destP += destStride;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
}
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
}
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
#endif
}
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -372,9 +401,16 @@ void vadd(const float* source1P,
float* destP,
int destStride,
size_t framesToProcess) {
- int n = framesToProcess;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if CPU(X86) && !defined(__SSE2__)
+ if (cpu.has_sse2()) {
+ blink::VectorMath::SSE2::vadd(source1P, sourceStride1, source2P, sourceStride2, destP, destStride, framesToProcess);
+ return;
+ }
+#endif
-#if CPU(X86) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+ int n = framesToProcess;
+
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
if ((sourceStride1 == 1) && (sourceStride2 == 1) && (destStride == 1)) {
// If the sourceP address is not 16-byte aligned, the first several frames
// (at most three) should be processed separately.
@@ -501,7 +537,7 @@ void vadd(const float* source1P,
source2P += sourceStride2;
destP += destStride;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
}
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
}
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
#endif
}
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -513,9 +549,16 @@ void vmul(const float* source1P,
float* destP,
int destStride,
size_t framesToProcess) {
- int n = framesToProcess;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if CPU(X86) && !defined(__SSE2__)
+ if (cpu.has_sse2()) {
+ blink::VectorMath::SSE2::vmul(source1P, sourceStride1, source2P, sourceStride2, destP, destStride, framesToProcess);
+ return;
+ }
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+
+int n = framesToProcess;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
if ((sourceStride1 == 1) && (sourceStride2 == 1) && (destStride == 1)) {
// If the source1P address is not 16-byte aligned, the first several frames
// (at most three) should be processed separately.
@@ -614,8 +657,15 @@ void zvmul(const float* real1P,
float* realDestP,
float* imagDestP,
size_t framesToProcess) {
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if CPU(X86) && !defined(__SSE2__)
+ if (cpu.has_sse2()) {
+ blink::VectorMath::SSE2::zvmul(real1P, imag1P, real2P, imag2P, realDestP, imagDestP, framesToProcess);
+ return;
+ }
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+
unsigned i = 0;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
// Only use the SSE optimization in the very common case that all addresses
// are 16-byte aligned. Otherwise, fall through to the scalar code below.
if (!(reinterpret_cast<uintptr_t>(real1P) & 0x0F) &&
@@ -671,10 +721,17 @@ void vsvesq(const float* sourceP,
int sourceStride,
float* sumP,
size_t framesToProcess) {
- int n = framesToProcess;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if CPU(X86) && !defined(__SSE2__)
+ if (cpu.has_sse2()) {
+ blink::VectorMath::SSE2::vsvesq(sourceP, sourceStride, sumP, framesToProcess);
+ return;
+ }
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
+
+ int n = framesToProcess;
float sum = 0;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
if (sourceStride == 1) {
// If the sourceP address is not 16-byte aligned, the first several frames
// (at most three) should be processed separately.
@@ -740,10 +797,16 @@ void vmaxmgv(const float* sourceP,
int sourceStride,
float* maxP,
size_t framesToProcess) {
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#if CPU(X86) && !defined(__SSE2__)
+ if (cpu.has_sse2()) {
+ blink::VectorMath::SSE2::vmaxmgv(sourceP, sourceStride, maxP, framesToProcess);
+ return;
+ }
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
int n = framesToProcess;
float max = 0;
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
if (sourceStride == 1) {
// If the sourceP address is not 16-byte aligned, the first several frames
// (at most three) should be processed separately.
@@ -839,6 +902,7 @@ void vclip(const float* sourceP,
float* destP,
int destStride,
size_t framesToProcess) {
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+#ifndef BUILD_ONLY_THE_SSE2_PARTS
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
int n = framesToProcess;
float lowThreshold = *lowThresholdP;
float highThreshold = *highThresholdP;
@@ -888,6 +952,11 @@ void vclip(const float* sourceP,
destP += destStride;
}
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
}
+#endif
+
+#ifdef BUILD_ONLY_THE_SSE2_PARTS
+} // namespace SSE2
+#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
#endif // OS(MACOSX)
diff --git a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/VectorMath.h b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/VectorMath.h
index 8f375071c..d92e5b85e 100644
--- a/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/VectorMath.h
+++ b/src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/VectorMath.h
@@ -27,6 +27,7 @@
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
#define VectorMath_h
#include "platform/PlatformExport.h"
+#include "wtf/CPU.h"
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
#include "wtf/build_config.h"
#include <cstddef>
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -97,6 +98,28 @@ PLATFORM_EXPORT void vclip(const float* sourceP,
int destStride,
size_t framesToProcess);
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
no-sse2.patch: Fix typos in the patched header files
2016-01-17 04:10:15 +00:00
+#if CPU(X86)
no-sse2.patch: fix audio/VectorMath and audio/SincResampler FTBFS * Fix src/3rdparty/chromium/third_party/WebKit/Source/platform/audio/ SincResampler.cpp and VectorMath.cpp to not require SSE2 on x86 (including runtime detection). * Also add runtime detection for DirectConvolver.cpp.
2016-01-17 03:30:41 +00:00
+namespace SSE2 {
+// Vector scalar multiply and then add.
+PLATFORM_EXPORT void vsma(const float* sourceP, int sourceStride, const float* scale, float* destP, int destStride, size_t framesToProcess);
+
+PLATFORM_EXPORT void vsmul(const float* sourceP, int sourceStride, const float* scale, float* destP, int destStride, size_t framesToProcess);
+PLATFORM_EXPORT void vadd(const float* source1P, int sourceStride1, const float* source2P, int sourceStride2, float* destP, int destStride, size_t framesToProcess);
+
+// Finds the maximum magnitude of a float vector.
+PLATFORM_EXPORT void vmaxmgv(const float* sourceP, int sourceStride, float* maxP, size_t framesToProcess);
+
+// Sums the squares of a float vector's elements.
+PLATFORM_EXPORT void vsvesq(const float* sourceP, int sourceStride, float* sumP, size_t framesToProcess);
+
+// For an element-by-element multiply of two float vectors.
+PLATFORM_EXPORT void vmul(const float* source1P, int sourceStride1, const float* source2P, int sourceStride2, float* destP, int destStride, size_t framesToProcess);
+
+// Multiplies two complex vectors.
+PLATFORM_EXPORT void zvmul(const float* real1P, const float* imag1P, const float* real2P, const float* imag2P, float* realDestP, float* imagDestP, size_t framesToProcess);
+}
+#endif
+
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
} // namespace VectorMath
} // namespace blink
diff --git a/src/3rdparty/chromium/third_party/WebKit/Source/platform/graphics/cpu/x86/WebGLImageConversionSSE.h b/src/3rdparty/chromium/third_party/WebKit/Source/platform/graphics/cpu/x86/WebGLImageConversionSSE.h
index e30fcb464..4b3437057 100644
--- a/src/3rdparty/chromium/third_party/WebKit/Source/platform/graphics/cpu/x86/WebGLImageConversionSSE.h
+++ b/src/3rdparty/chromium/third_party/WebKit/Source/platform/graphics/cpu/x86/WebGLImageConversionSSE.h
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
@@ -5,7 +5,7 @@
#ifndef WebGLImageConversionSSE_h
#define WebGLImageConversionSSE_h
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
#include <emmintrin.h>
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/third_party/WebKit/Source/platform/graphics/gpu/WebGLImageConversion.cpp b/src/3rdparty/chromium/third_party/WebKit/Source/platform/graphics/gpu/WebGLImageConversion.cpp
index be417ea88..e79034424 100644
--- a/src/3rdparty/chromium/third_party/WebKit/Source/platform/graphics/gpu/WebGLImageConversion.cpp
+++ b/src/3rdparty/chromium/third_party/WebKit/Source/platform/graphics/gpu/WebGLImageConversion.cpp
@@ -441,7 +441,7 @@ void unpack<WebGLImageConversion::DataFormatBGRA8, uint8_t, uint8_t>(
const uint32_t* source32 = reinterpret_cast_ptr<const uint32_t*>(source);
uint32_t* destination32 = reinterpret_cast_ptr<uint32_t*>(destination);
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
SIMD::unpackOneRowOfBGRA8LittleToRGBA8(source32, destination32, pixelsPerRow);
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
#if HAVE(MIPS_MSA_INTRINSICS)
@@ -467,7 +467,7 @@ void unpack<WebGLImageConversion::DataFormatRGBA5551, uint16_t, uint8_t>(
const uint16_t* source,
uint8_t* destination,
unsigned pixelsPerRow) {
no-sse2 patch: Fix conditionals in 3 more sections of WebGLImageConversion.cpp
2016-07-18 20:53:48 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
SIMD::unpackOneRowOfRGBA5551LittleToRGBA8(source, destination, pixelsPerRow);
no-sse2 patch: Fix conditionals in 3 more sections of WebGLImageConversion.cpp
2016-07-18 20:53:48 +00:00
#endif
#if HAVE(ARM_NEON_INTRINSICS)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -496,7 +496,7 @@ void unpack<WebGLImageConversion::DataFormatRGBA4444, uint16_t, uint8_t>(
const uint16_t* source,
uint8_t* destination,
unsigned pixelsPerRow) {
no-sse2 patch: Fix conditionals in 3 more sections of WebGLImageConversion.cpp
2016-07-18 20:53:48 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
SIMD::unpackOneRowOfRGBA4444LittleToRGBA8(source, destination, pixelsPerRow);
no-sse2 patch: Fix conditionals in 3 more sections of WebGLImageConversion.cpp
2016-07-18 20:53:48 +00:00
#endif
#if HAVE(ARM_NEON_INTRINSICS)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -711,7 +711,7 @@ void pack<WebGLImageConversion::DataFormatR8,
uint8_t>(const uint8_t* source,
uint8_t* destination,
unsigned pixelsPerRow) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
SIMD::packOneRowOfRGBA8LittleToR8(source, destination, pixelsPerRow);
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
#if HAVE(MIPS_MSA_INTRINSICS)
@@ -768,7 +768,7 @@ void pack<WebGLImageConversion::DataFormatRA8,
uint8_t>(const uint8_t* source,
uint8_t* destination,
unsigned pixelsPerRow) {
no-sse2 patch: Fix conditionals in 3 more sections of WebGLImageConversion.cpp
2016-07-18 20:53:48 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
SIMD::packOneRowOfRGBA8LittleToRA8(source, destination, pixelsPerRow);
no-sse2 patch: Fix conditionals in 3 more sections of WebGLImageConversion.cpp
2016-07-18 20:53:48 +00:00
#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
#if HAVE(MIPS_MSA_INTRINSICS)
@@ -880,7 +880,7 @@ void pack<WebGLImageConversion::DataFormatRGBA8,
uint8_t>(const uint8_t* source,
uint8_t* destination,
unsigned pixelsPerRow) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
-#if CPU(X86) || CPU(X86_64)
+#if (CPU(X86) && defined(__SSE2__)) || CPU(X86_64)
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
SIMD::packOneRowOfRGBA8LittleToRGBA8(source, destination, pixelsPerRow);
#endif
#if HAVE(MIPS_MSA_INTRINSICS)
diff --git a/src/3rdparty/chromium/third_party/qcms/BUILD.gn b/src/3rdparty/chromium/third_party/qcms/BUILD.gn
index 1183569be..771dd02c8 100644
--- a/src/3rdparty/chromium/third_party/qcms/BUILD.gn
+++ b/src/3rdparty/chromium/third_party/qcms/BUILD.gn
@@ -30,8 +30,8 @@ static_library("qcms") {
]
if (current_cpu == "x86" || current_cpu == "x64") {
- defines = [ "SSE2_ENABLE" ]
- sources += [ "src/transform-sse2.c" ]
+ defines = [ "SSE2_ENABLE" ] # runtime detection
+ deps = [ "qcms_sse2" ]
}
}
@@ -74,3 +74,16 @@ if (!disable_qcms) {
public_configs = [ ":qcms_config" ]
}
}
+
+source_set("qcms_sse2") {
+ configs -= [ "//build/config/compiler:chromium_code" ]
+ configs += [ "//build/config/compiler:no_chromium_code" ]
+ public_configs = [ ":qcms_config" ]
+
+ if (current_cpu == "x86" || current_cpu == "x64") {
+ defines = [ "SSE2_ENABLE" ]
+ sources = [ "src/transform-sse2.c" ]
+ cflags = [ "-msse2" ]
+ }
+}
+
diff --git a/src/3rdparty/chromium/third_party/webrtc/common_audio/real_fourier.cc b/src/3rdparty/chromium/third_party/webrtc/common_audio/real_fourier.cc
index e721346f3..ef5a4e4a7 100644
--- a/src/3rdparty/chromium/third_party/webrtc/common_audio/real_fourier.cc
+++ b/src/3rdparty/chromium/third_party/webrtc/common_audio/real_fourier.cc
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
@@ -14,6 +14,7 @@
#include "webrtc/common_audio/real_fourier_ooura.h"
#include "webrtc/common_audio/real_fourier_openmax.h"
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
no-sse2 patch: Fix/update the location of cpu_features_wrapper.h It moved from "webrtc/system_wrappers/interface/cpu_features_wrapper.h" to "webrtc/system_wrappers/include/cpu_features_wrapper.h".
2016-07-18 17:28:55 +00:00
+#include "webrtc/system_wrappers/include/cpu_features_wrapper.h"
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
namespace webrtc {
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -23,7 +24,15 @@ const size_t RealFourier::kFftBufferAlignment = 32;
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
std::unique_ptr<RealFourier> RealFourier::Create(int fft_order) {
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#if defined(RTC_USE_OPENMAX_DL)
+#if defined(WEBRTC_ARCH_X86_FAMILY) && !defined(__SSE2__)
+ // x86 CPU detection required.
+ if (WebRtc_GetCPUInfo(kSSE2))
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ return std::unique_ptr<RealFourier>(new RealFourierOpenmax(fft_order));
no-sse2 patch: Fix FFT (RealFourier) in webrtc on non-SSE2 x86 * Tue Feb 23 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.0-0.18.beta - no-sse2 patch: Fix FFT (RealFourier) in webrtc on non-SSE2 x86
2016-02-24 03:50:18 +00:00
+ else
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ return std::unique_ptr<RealFourier>(new RealFourierOoura(fft_order));
no-sse2 patch: Fix FFT (RealFourier) in webrtc on non-SSE2 x86 * Tue Feb 23 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.0-0.18.beta - no-sse2 patch: Fix FFT (RealFourier) in webrtc on non-SSE2 x86
2016-02-24 03:50:18 +00:00
+#else
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
return std::unique_ptr<RealFourier>(new RealFourierOpenmax(fft_order));
no-sse2 patch: Fix FFT (RealFourier) in webrtc on non-SSE2 x86 * Tue Feb 23 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.0-0.18.beta - no-sse2 patch: Fix FFT (RealFourier) in webrtc on non-SSE2 x86
2016-02-24 03:50:18 +00:00
+#endif
Do not require SSE2 on i686 See https://bugzilla.redhat.com/show_bug.cgi?id=1244196#c22 for the history of the patch. Compared to the last revision there (v5): * I fixed WebGLImageConversion.cpp to not assume SSE2 on x86. Unfortunately, it would be really hard to make runtime detection work there, because it uses intrinsics in inline templates, and GCC only supports intrinsics when building the whole code for the vector instruction set. So I had to turn off that SSE2 code entirely when __SSE2__ is not defined at compile time. * I fixed webrtc to check for SSE2 at runtime (using the runtime checking code that webrtc already has and uses in several places) before attempting to use OpenMAX on x86. OpenMAX requires SSE2. So we just fall back to the Ooura implementation of FFTs when needed. TODO: build V8 shared and twice (once for x87, once for SSE2)
2016-01-16 23:17:04 +00:00
#else
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
return std::unique_ptr<RealFourier>(new RealFourierOoura(fft_order));
Fix no-sse2 patch. The patch tool insists on symmetric context for some reason.
2016-02-24 13:08:39 +00:00
#endif
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/v8/BUILD.gn b/src/3rdparty/chromium/v8/BUILD.gn
index 9b209dfff..d88261d53 100644
--- a/src/3rdparty/chromium/v8/BUILD.gn
+++ b/src/3rdparty/chromium/v8/BUILD.gn
@@ -1769,43 +1769,41 @@ v8_static_library("v8_base") {
if (v8_current_cpu == "x86") {
sources += [ ### gcmole(arch:ia32) ###
- "src/builtins/ia32/builtins-ia32.cc",
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
- "src/compiler/ia32/code-generator-ia32.cc",
- "src/compiler/ia32/instruction-codes-ia32.h",
- "src/compiler/ia32/instruction-scheduler-ia32.cc",
- "src/compiler/ia32/instruction-selector-ia32.cc",
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
- "src/crankshaft/ia32/lithium-codegen-ia32.cc",
- "src/crankshaft/ia32/lithium-codegen-ia32.h",
- "src/crankshaft/ia32/lithium-gap-resolver-ia32.cc",
- "src/crankshaft/ia32/lithium-gap-resolver-ia32.h",
- "src/crankshaft/ia32/lithium-ia32.cc",
- "src/crankshaft/ia32/lithium-ia32.h",
- "src/debug/ia32/debug-ia32.cc",
- "src/full-codegen/ia32/full-codegen-ia32.cc",
no-sse2.patch: Fix typos * Fix typo in blink_platform.gyp. * Restore the v8 parts that accidentally went missing.
2016-01-17 03:46:25 +00:00
- "src/ia32/assembler-ia32-inl.h",
- "src/ia32/assembler-ia32.cc",
- "src/ia32/assembler-ia32.h",
- "src/ia32/code-stubs-ia32.cc",
- "src/ia32/code-stubs-ia32.h",
- "src/ia32/codegen-ia32.cc",
- "src/ia32/codegen-ia32.h",
- "src/ia32/cpu-ia32.cc",
- "src/ia32/deoptimizer-ia32.cc",
- "src/ia32/disasm-ia32.cc",
- "src/ia32/frames-ia32.cc",
- "src/ia32/frames-ia32.h",
- "src/ia32/interface-descriptors-ia32.cc",
- "src/ia32/macro-assembler-ia32.cc",
- "src/ia32/macro-assembler-ia32.h",
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
- "src/ia32/simulator-ia32.cc",
- "src/ia32/simulator-ia32.h",
no-sse2.patch: Fix typos * Fix typo in blink_platform.gyp. * Restore the v8 parts that accidentally went missing.
2016-01-17 03:46:25 +00:00
- "src/ic/ia32/access-compiler-ia32.cc",
- "src/ic/ia32/handler-compiler-ia32.cc",
- "src/ic/ia32/ic-compiler-ia32.cc",
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
- "src/ic/ia32/ic-ia32.cc",
no-sse2.patch: Fix typos * Fix typo in blink_platform.gyp. * Restore the v8 parts that accidentally went missing.
2016-01-17 03:46:25 +00:00
- "src/ic/ia32/stub-cache-ia32.cc",
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
- "src/regexp/ia32/regexp-macro-assembler-ia32.cc",
- "src/regexp/ia32/regexp-macro-assembler-ia32.h",
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ "src/compiler/x87/code-generator-x87.cc",
+ "src/compiler/x87/instruction-codes-x87.h",
+ "src/compiler/x87/instruction-scheduler-x87.cc",
+ "src/compiler/x87/instruction-selector-x87.cc",
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ "src/crankshaft/x87/lithium-codegen-x87.cc",
+ "src/crankshaft/x87/lithium-codegen-x87.h",
+ "src/crankshaft/x87/lithium-gap-resolver-x87.cc",
+ "src/crankshaft/x87/lithium-gap-resolver-x87.h",
+ "src/crankshaft/x87/lithium-x87.cc",
+ "src/crankshaft/x87/lithium-x87.h",
+ "src/debug/x87/debug-x87.cc",
+ "src/full-codegen/x87/full-codegen-x87.cc",
no-sse2.patch: Fix typos * Fix typo in blink_platform.gyp. * Restore the v8 parts that accidentally went missing.
2016-01-17 03:46:25 +00:00
+ "src/x87/assembler-x87-inl.h",
+ "src/x87/assembler-x87.cc",
+ "src/x87/assembler-x87.h",
+ "src/x87/builtins-x87.cc",
+ "src/x87/code-stubs-x87.cc",
+ "src/x87/code-stubs-x87.h",
+ "src/x87/codegen-x87.cc",
+ "src/x87/codegen-x87.h",
+ "src/x87/cpu-x87.cc",
+ "src/x87/deoptimizer-x87.cc",
+ "src/x87/disasm-x87.cc",
+ "src/x87/frames-x87.cc",
+ "src/x87/frames-x87.h",
+ "src/x87/interface-descriptors-x87.cc",
+ "src/x87/macro-assembler-x87.cc",
+ "src/x87/macro-assembler-x87.h",
+ "src/ic/x87/access-compiler-x87.cc",
+ "src/ic/x87/handler-compiler-x87.cc",
+ "src/ic/x87/ic-compiler-x87.cc",
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ "src/ic/x87/ic-x87.cc",
no-sse2.patch: Fix typos * Fix typo in blink_platform.gyp. * Restore the v8 parts that accidentally went missing.
2016-01-17 03:46:25 +00:00
+ "src/ic/x87/stub-cache-x87.cc",
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ "src/regexp/x87/regexp-macro-assembler-x87.cc",
+ "src/regexp/x87/regexp-macro-assembler-x87.h",
no-sse2.patch: Fix typos * Fix typo in blink_platform.gyp. * Restore the v8 parts that accidentally went missing.
2016-01-17 03:46:25 +00:00
]
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
} else if (v8_current_cpu == "x64") {
sources += [ ### gcmole(arch:x64) ###
diff --git a/src/3rdparty/chromium/v8/src/make-v8-sse2-gyp.sh b/src/3rdparty/chromium/v8/src/make-v8-sse2-gyp.sh
new file mode 100644
index 000000000..0e7ec1621
--- /dev/null
+++ b/src/3rdparty/chromium/v8/src/make-v8-sse2-gyp.sh
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
@@ -0,0 +1,56 @@
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+#!/bin/sh
+# This script renames the v8 targets to _sse2 names so that they do not conflict
+# with the non-SSE2 versions.
+
+# Copyright 2016 Kevin Kofler. All rights reserved.
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+# * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+# * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following
+# disclaimer in the documentation and/or other materials provided
+# with the distribution.
+# * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived
+# from this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+# add comment noting that the file is generated
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+echo "# Generated from v8.gyp by make-v8-sse2-gyp.sh" >v8_sse2.gyp
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+# rename all target names
+SUBTARGETS=`grep "'target_name': '" v8.gyp | sed -e "s/^.*'target_name': '//g" -e "s/',$//g"`
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+SEDS=
+for SUBTARGET in $SUBTARGETS ; do
no-sse2.patch: Fix another error in v8_sse2.gyp (by fixing make-v8-sse2-gyp.sh)
2016-01-19 03:11:36 +00:00
+ SEDS=$SEDS\ -e\ "s/'$SUBTARGET\(['#]\)/'${SUBTARGET}_sse2\1/g"
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+done
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+# in addition:
+# * set v8_target_arch to "ia32" (instead of "x87")
+# * rename all actions
+# * fix mksnapshot_exec to match the renamed target
+# * rename the generated snapshot.cc (but not mksnapshot.cc) to snapshot_sse2.cc
+# * rename the generated *libraries.cc to *libraries_sse2.cc
+# * rename the generated *.bin to *_sse2.bin
+# * set product_name and product_dir for the v8_sse2 target
no-sse2.patch: Fix invalid gyp syntax in v8_sse2.gyp (by fixing make-v8-sse2-gyp.sh)
2016-01-19 02:52:05 +00:00
+sed -e "s/^\( 'variables': {\)/\1\n 'v8_target_arch': 'ia32',/g" \
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ -e "s/\('action_name': '\)/\1v8_sse2_/g" \
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ $SEDS \
+ -e "s/\('mksnapshot_exec': '.*mksnapshot\)/\1_sse2/g" \
+ -e "s#/snapshot\.cc#/snapshot_sse2.cc#g" \
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ -e "s/libraries\.cc/libraries_sse2.cc/g" \
+ -e "s/\.bin/_sse2.bin/g" \
no-sse2.patch: Fix the output path in v8_sse2.gyp (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 04:54:16 +00:00
+ -e "s#^\( *\)\('target_name': 'v8_sse2',\)#\1\2\n\1'product_name': 'v8',\n\1'product_dir': '<(PRODUCT_DIR)/lib/sse2',#g" \
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ v8.gyp >>v8_sse2.gyp
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/3rdparty/chromium/v8/src/v8.gyp b/src/3rdparty/chromium/v8/src/v8.gyp
index 020ec0928..2add932fc 100644
--- a/src/3rdparty/chromium/v8/src/v8.gyp
+++ b/src/3rdparty/chromium/v8/src/v8.gyp
@@ -51,8 +51,8 @@
no-sse2.patch: Build V8 as a shared library on i686 * Mon Jan 18 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.0-0.14.beta.nosse2.1 - Build V8 as a shared library on i686 to allow for swappable backends This is the first prerequisite step towards building both the x87 and SSE2 backends.
2016-01-18 00:10:42 +00:00
}, {
'toolsets': ['target'],
}],
- ['component=="shared_library"', {
- 'type': '<(component)',
+ ['v8_component=="shared_library"', {
+ 'type': '<(v8_component)',
'sources': [
# Note: on non-Windows we still build this file so that gyp
# has some sources to link into the component.
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -151,7 +151,7 @@
no-sse2.patch: Build V8 as a shared library on i686 * Mon Jan 18 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.0-0.14.beta.nosse2.1 - Build V8 as a shared library on i686 to allow for swappable backends This is the first prerequisite step towards building both the x87 and SSE2 backends.
2016-01-18 00:10:42 +00:00
'js2c',
],
}],
- ['component=="shared_library"', {
+ ['v8_component=="shared_library"', {
'defines': [
'BUILDING_V8_SHARED',
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
],
@@ -243,7 +243,7 @@
no-sse2.patch: Build V8 as a shared library on i686 * Mon Jan 18 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.0-0.14.beta.nosse2.1 - Build V8 as a shared library on i686 to allow for swappable backends This is the first prerequisite step towards building both the x87 and SSE2 backends.
2016-01-18 00:10:42 +00:00
'toolsets': ['target'],
'dependencies': ['js2c'],
}],
- ['component=="shared_library"', {
+ ['v8_component=="shared_library"', {
'defines': [
'BUILDING_V8_SHARED',
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
],
@@ -270,7 +270,7 @@
no-sse2.patch: Build V8 as a shared library on i686 * Mon Jan 18 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.0-0.14.beta.nosse2.1 - Build V8 as a shared library on i686 to allow for swappable backends This is the first prerequisite step towards building both the x87 and SSE2 backends.
2016-01-18 00:10:42 +00:00
'natives_blob',
],
}],
- ['component=="shared_library"', {
+ ['v8_component=="shared_library"', {
'defines': [
'BUILDING_V8_SHARED',
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
],
@@ -1727,7 +1727,7 @@
no-sse2.patch: Build V8 as a shared library on i686 * Mon Jan 18 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.0-0.14.beta.nosse2.1 - Build V8 as a shared library on i686 to allow for swappable backends This is the first prerequisite step towards building both the x87 and SSE2 backends.
2016-01-18 00:10:42 +00:00
# See http://crbug.com/485155.
'msvs_shard': 4,
}],
- ['component=="shared_library"', {
+ ['v8_component=="shared_library"', {
'defines': [
'BUILDING_V8_SHARED',
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
],
diff --git a/src/3rdparty/chromium/v8/src/v8_sse2.gyp b/src/3rdparty/chromium/v8/src/v8_sse2.gyp
new file mode 100644
index 000000000..140b071a6
--- /dev/null
+++ b/src/3rdparty/chromium/v8/src/v8_sse2.gyp
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
@@ -0,0 +1,2313 @@
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+# Generated from v8.gyp by make-v8-sse2-gyp.sh
+# Copyright 2012 the V8 project authors. All rights reserved.
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+# * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+# * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following
+# disclaimer in the documentation and/or other materials provided
+# with the distribution.
+# * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived
+# from this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+{
+ 'variables': {
no-sse2.patch: Fix invalid gyp syntax in v8_sse2.gyp (by fixing make-v8-sse2-gyp.sh)
2016-01-19 02:52:05 +00:00
+ 'v8_target_arch': 'ia32',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'v8_code': 1,
+ 'v8_random_seed%': 314159265,
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ 'v8_vector_stores%': 0,
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'embed_script%': "",
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'warmup_script%': "",
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'v8_extra_library_files%': [],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ 'v8_experimental_extra_library_files%': [],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'mksnapshot_exec': '<(PRODUCT_DIR)/<(EXECUTABLE_PREFIX)mksnapshot_sse2<(EXECUTABLE_SUFFIX)',
+ 'remove_v8base_debug_symbols%': 0,
+ 'conditions': [
+ # build V8 shared on ia32 so we can swap x87 vs. SSE2 builds
+ ['target_arch == "ia32"', {
+ 'v8_component%': 'shared_library',
+ }, {
+ 'v8_component%': '<(component)',
+ }],
+ ],
+ },
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'includes': ['../gypfiles/toolchain.gypi', '../gypfiles/features.gypi'],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'targets': [
+ {
no-sse2.patch: Fix invalid gyp syntax in v8_sse2.gyp (by fixing make-v8-sse2-gyp.sh)
2016-01-19 02:52:05 +00:00
+ 'target_name': 'v8_sse2',
+ 'product_name': 'v8',
no-sse2.patch: Fix the output path in v8_sse2.gyp (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 04:54:16 +00:00
+ 'product_dir': '<(PRODUCT_DIR)/lib/sse2',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'dependencies_traverse': 1,
+ 'dependencies': ['v8_maybe_snapshot_sse2'],
+ 'conditions': [
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host', 'target'],
+ }, {
+ 'toolsets': ['target'],
+ }],
+ ['v8_component=="shared_library"', {
+ 'type': '<(v8_component)',
+ 'sources': [
+ # Note: on non-Windows we still build this file so that gyp
+ # has some sources to link into the component.
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'v8dll-main.cc',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'include_dirs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '..',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'defines': [
+ 'V8_SHARED',
+ 'BUILDING_V8_SHARED',
+ ],
+ 'direct_dependent_settings': {
+ 'defines': [
+ 'V8_SHARED',
+ 'USING_V8_SHARED',
+ ],
+ },
+ 'target_conditions': [
+ ['OS=="android" and _toolset=="target"', {
+ 'libraries': [
+ '-llog',
+ ],
+ 'include_dirs': [
+ 'src/common/android/include',
+ ],
+ }],
+ ],
+ 'conditions': [
+ ['OS=="mac"', {
+ 'xcode_settings': {
+ 'OTHER_LDFLAGS': ['-dynamiclib', '-all_load']
+ },
+ }],
+ ['soname_version!=""', {
+ 'product_extension': 'so.<(soname_version)',
+ }],
+ ],
+ },
+ {
+ 'type': 'none',
+ }],
+ ],
+ 'direct_dependent_settings': {
+ 'include_dirs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../include',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ },
+ },
+ {
+ # This rule delegates to either v8_snapshot, v8_nosnapshot, or
+ # v8_external_snapshot, depending on the current variables.
+ # The intention is to make the 'calling' rules a bit simpler.
+ 'target_name': 'v8_maybe_snapshot_sse2',
+ 'type': 'none',
+ 'conditions': [
+ ['v8_use_snapshot!="true"', {
+ # The dependency on v8_base should come from a transitive
+ # dependency however the Android toolchain requires libv8_base.a
+ # to appear before libv8_snapshot.a so it's listed explicitly.
+ 'dependencies': ['v8_base_sse2', 'v8_nosnapshot_sse2'],
+ }],
+ ['v8_use_snapshot=="true" and v8_use_external_startup_data==0', {
+ # The dependency on v8_base should come from a transitive
+ # dependency however the Android toolchain requires libv8_base.a
+ # to appear before libv8_snapshot.a so it's listed explicitly.
+ 'dependencies': ['v8_base_sse2', 'v8_snapshot_sse2'],
+ }],
+ ['v8_use_snapshot=="true" and v8_use_external_startup_data==1 and want_separate_host_toolset==0', {
+ 'dependencies': ['v8_base_sse2', 'v8_external_snapshot_sse2'],
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'inputs': [ '<(PRODUCT_DIR)/snapshot_blob_sse2.bin', ],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ }],
+ ['v8_use_snapshot=="true" and v8_use_external_startup_data==1 and want_separate_host_toolset==1', {
+ 'dependencies': ['v8_base_sse2', 'v8_external_snapshot_sse2'],
+ 'target_conditions': [
+ ['_toolset=="host"', {
+ 'inputs': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(PRODUCT_DIR)/snapshot_blob_host_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }, {
+ 'inputs': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(PRODUCT_DIR)/snapshot_blob_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ],
+ }],
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host', 'target'],
+ }, {
+ 'toolsets': ['target'],
+ }],
+ ]
+ },
+ {
+ 'target_name': 'v8_snapshot_sse2',
+ 'type': 'static_library',
+ 'conditions': [
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host', 'target'],
+ 'dependencies': [
no-sse2.patch: Fix another error in v8_sse2.gyp (by fixing make-v8-sse2-gyp.sh)
2016-01-19 03:11:36 +00:00
+ 'mksnapshot_sse2#host',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'js2c_sse2#host',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }, {
+ 'toolsets': ['target'],
+ 'dependencies': [
+ 'mksnapshot_sse2',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'js2c_sse2',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['v8_component=="shared_library"', {
+ 'defines': [
+ 'V8_SHARED',
+ 'BUILDING_V8_SHARED',
+ ],
+ 'direct_dependent_settings': {
+ 'defines': [
+ 'V8_SHARED',
+ 'USING_V8_SHARED',
+ ],
+ },
+ }],
+ ],
+ 'dependencies': [
+ 'v8_base_sse2',
+ ],
+ 'include_dirs+': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '..',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'sources': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/libraries_sse2.cc',
+ '<(SHARED_INTERMEDIATE_DIR)/experimental-libraries_sse2.cc',
+ '<(SHARED_INTERMEDIATE_DIR)/extras-libraries_sse2.cc',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/experimental-extras-libraries_sse2.cc',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ '<(INTERMEDIATE_DIR)/snapshot_sse2.cc',
+ ],
+ 'actions': [
+ {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'action_name': 'v8_sse2_run_mksnapshot',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'inputs': [
+ '<(mksnapshot_exec)',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ ],
+ 'conditions': [
+ ['embed_script!=""', {
+ 'inputs': [
+ '<(embed_script)',
+ ],
+ }],
+ ['warmup_script!=""', {
+ 'inputs': [
+ '<(warmup_script)',
+ ],
+ }],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'outputs': [
+ '<(INTERMEDIATE_DIR)/snapshot_sse2.cc',
+ ],
+ 'variables': {
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'mksnapshot_flags': [],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'conditions': [
+ ['v8_random_seed!=0', {
+ 'mksnapshot_flags': ['--random-seed', '<(v8_random_seed)'],
+ }],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ ['v8_vector_stores!=0', {
+ 'mksnapshot_flags': ['--vector-stores'],
+ }],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ },
+ 'action': [
+ '<(mksnapshot_exec)',
+ '<@(mksnapshot_flags)',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '--startup_src', '<@(INTERMEDIATE_DIR)/snapshot_sse2.cc',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ '<(embed_script)',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '<(warmup_script)',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ },
+ ],
+ },
+ {
+ 'target_name': 'v8_nosnapshot_sse2',
+ 'type': 'static_library',
+ 'dependencies': [
+ 'v8_base_sse2',
+ ],
+ 'include_dirs+': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '..',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'sources': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/libraries_sse2.cc',
+ '<(SHARED_INTERMEDIATE_DIR)/experimental-libraries_sse2.cc',
+ '<(SHARED_INTERMEDIATE_DIR)/extras-libraries_sse2.cc',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/experimental-extras-libraries_sse2.cc',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'snapshot/snapshot-empty.cc',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'conditions': [
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host', 'target'],
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'dependencies': ['js2c_sse2#host'],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ }, {
+ 'toolsets': ['target'],
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'dependencies': ['js2c_sse2'],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ }],
+ ['v8_component=="shared_library"', {
+ 'defines': [
+ 'BUILDING_V8_SHARED',
+ 'V8_SHARED',
+ ],
+ }],
+ ]
+ },
+ {
+ 'target_name': 'v8_external_snapshot_sse2',
+ 'type': 'static_library',
+ 'conditions': [
+ [ 'v8_use_external_startup_data==1', {
+ 'conditions': [
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host', 'target'],
+ 'dependencies': [
no-sse2.patch: Fix another error in v8_sse2.gyp (by fixing make-v8-sse2-gyp.sh)
2016-01-19 03:11:36 +00:00
+ 'mksnapshot_sse2#host',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'js2c_sse2#host',
+ 'natives_blob_sse2',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ]}, {
+ 'toolsets': ['target'],
+ 'dependencies': [
+ 'mksnapshot_sse2',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'js2c_sse2',
+ 'natives_blob_sse2',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['v8_component=="shared_library"', {
+ 'defines': [
+ 'V8_SHARED',
+ 'BUILDING_V8_SHARED',
+ ],
+ 'direct_dependent_settings': {
+ 'defines': [
+ 'V8_SHARED',
+ 'USING_V8_SHARED',
+ ],
+ },
+ }],
+ ],
+ 'dependencies': [
+ 'v8_base_sse2',
+ ],
+ 'include_dirs+': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '..',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'snapshot/natives-external.cc',
+ 'snapshot/snapshot-external.cc',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'actions': [
+ {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'action_name': 'v8_sse2_run_mksnapshot (external)',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'inputs': [
+ '<(mksnapshot_exec)',
+ ],
+ 'variables': {
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'mksnapshot_flags': [],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'conditions': [
+ ['v8_random_seed!=0', {
+ 'mksnapshot_flags': ['--random-seed', '<(v8_random_seed)'],
+ }],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ ['v8_vector_stores!=0', {
+ 'mksnapshot_flags': ['--vector-stores'],
+ }],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ },
+ 'conditions': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ ['embed_script!=""', {
+ 'inputs': [
+ '<(embed_script)',
+ ],
+ }],
+ ['warmup_script!=""', {
+ 'inputs': [
+ '<(warmup_script)',
+ ],
+ }],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ['want_separate_host_toolset==1', {
+ 'target_conditions': [
+ ['_toolset=="host"', {
+ 'outputs': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(PRODUCT_DIR)/snapshot_blob_host_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'action': [
+ '<(mksnapshot_exec)',
+ '<@(mksnapshot_flags)',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '--startup_blob', '<(PRODUCT_DIR)/snapshot_blob_host_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ '<(embed_script)',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '<(warmup_script)',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }, {
+ 'outputs': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(PRODUCT_DIR)/snapshot_blob_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'action': [
+ '<(mksnapshot_exec)',
+ '<@(mksnapshot_flags)',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '--startup_blob', '<(PRODUCT_DIR)/snapshot_blob_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ '<(embed_script)',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '<(warmup_script)',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ],
+ }, {
+ 'outputs': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(PRODUCT_DIR)/snapshot_blob_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'action': [
+ '<(mksnapshot_exec)',
+ '<@(mksnapshot_flags)',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '--startup_blob', '<(PRODUCT_DIR)/snapshot_blob_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ '<(embed_script)',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '<(warmup_script)',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ],
+ },
+ ],
+ }],
+ ],
+ },
+ {
+ 'target_name': 'v8_base_sse2',
+ 'type': 'static_library',
+ 'dependencies': [
+ 'v8_libbase_sse2',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'v8_libsampler_sse2',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'variables': {
+ 'optimize': 'max',
+ },
+ 'include_dirs+': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '..',
+ '<(DEPTH)',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'sources': [ ### gcmole(all) ###
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../include/v8-debug.h',
+ '../include/v8-experimental.h',
+ '../include/v8-platform.h',
+ '../include/v8-profiler.h',
+ '../include/v8-testing.h',
+ '../include/v8-util.h',
+ '../include/v8-version.h',
+ '../include/v8.h',
+ '../include/v8config.h',
+ 'accessors.cc',
+ 'accessors.h',
+ 'address-map.cc',
+ 'address-map.h',
+ 'allocation.cc',
+ 'allocation.h',
+ 'allocation-site-scopes.cc',
+ 'allocation-site-scopes.h',
+ 'api-experimental.cc',
+ 'api-experimental.h',
+ 'api.cc',
+ 'api.h',
+ 'api-arguments-inl.h',
+ 'api-arguments.cc',
+ 'api-arguments.h',
+ 'api-natives.cc',
+ 'api-natives.h',
+ 'arguments.cc',
+ 'arguments.h',
+ 'assembler.cc',
+ 'assembler.h',
+ 'assert-scope.h',
+ 'assert-scope.cc',
+ 'ast/ast-expression-rewriter.cc',
+ 'ast/ast-expression-rewriter.h',
+ 'ast/ast-expression-visitor.cc',
+ 'ast/ast-expression-visitor.h',
+ 'ast/ast-literal-reindexer.cc',
+ 'ast/ast-literal-reindexer.h',
+ 'ast/ast-numbering.cc',
+ 'ast/ast-numbering.h',
+ 'ast/ast-type-bounds.h',
+ 'ast/ast-value-factory.cc',
+ 'ast/ast-value-factory.h',
+ 'ast/ast.cc',
+ 'ast/ast.h',
+ 'ast/modules.cc',
+ 'ast/modules.h',
+ 'ast/prettyprinter.cc',
+ 'ast/prettyprinter.h',
+ 'ast/scopeinfo.cc',
+ 'ast/scopeinfo.h',
+ 'ast/scopes.cc',
+ 'ast/scopes.h',
+ 'ast/variables.cc',
+ 'ast/variables.h',
+ 'background-parsing-task.cc',
+ 'background-parsing-task.h',
+ 'bailout-reason.cc',
+ 'bailout-reason.h',
+ 'basic-block-profiler.cc',
+ 'basic-block-profiler.h',
+ 'bignum-dtoa.cc',
+ 'bignum-dtoa.h',
+ 'bignum.cc',
+ 'bignum.h',
+ 'bit-vector.cc',
+ 'bit-vector.h',
+ 'bootstrapper.cc',
+ 'bootstrapper.h',
+ 'builtins.cc',
+ 'builtins.h',
+ 'cached-powers.cc',
+ 'cached-powers.h',
+ 'cancelable-task.cc',
+ 'cancelable-task.h',
+ 'char-predicates.cc',
+ 'char-predicates-inl.h',
+ 'char-predicates.h',
+ 'checks.h',
+ 'code-events.h',
+ 'code-factory.cc',
+ 'code-factory.h',
+ 'code-stub-assembler.cc',
+ 'code-stub-assembler.h',
+ 'code-stubs.cc',
+ 'code-stubs.h',
+ 'code-stubs-hydrogen.cc',
+ 'codegen.cc',
+ 'codegen.h',
+ 'collector.h',
+ 'compilation-cache.cc',
+ 'compilation-cache.h',
+ 'compilation-dependencies.cc',
+ 'compilation-dependencies.h',
+ 'compilation-statistics.cc',
+ 'compilation-statistics.h',
+ 'compiler/access-builder.cc',
+ 'compiler/access-builder.h',
+ 'compiler/access-info.cc',
+ 'compiler/access-info.h',
+ 'compiler/all-nodes.cc',
+ 'compiler/all-nodes.h',
+ 'compiler/ast-graph-builder.cc',
+ 'compiler/ast-graph-builder.h',
+ 'compiler/ast-loop-assignment-analyzer.cc',
+ 'compiler/ast-loop-assignment-analyzer.h',
+ 'compiler/basic-block-instrumentor.cc',
+ 'compiler/basic-block-instrumentor.h',
+ 'compiler/branch-elimination.cc',
+ 'compiler/branch-elimination.h',
+ 'compiler/bytecode-branch-analysis.cc',
+ 'compiler/bytecode-branch-analysis.h',
+ 'compiler/bytecode-graph-builder.cc',
+ 'compiler/bytecode-graph-builder.h',
+ 'compiler/c-linkage.cc',
+ 'compiler/checkpoint-elimination.cc',
+ 'compiler/checkpoint-elimination.h',
+ 'compiler/code-generator-impl.h',
+ 'compiler/code-generator.cc',
+ 'compiler/code-generator.h',
+ 'compiler/code-assembler.cc',
+ 'compiler/code-assembler.h',
+ 'compiler/common-node-cache.cc',
+ 'compiler/common-node-cache.h',
+ 'compiler/common-operator-reducer.cc',
+ 'compiler/common-operator-reducer.h',
+ 'compiler/common-operator.cc',
+ 'compiler/common-operator.h',
+ 'compiler/control-builders.cc',
+ 'compiler/control-builders.h',
+ 'compiler/control-equivalence.cc',
+ 'compiler/control-equivalence.h',
+ 'compiler/control-flow-optimizer.cc',
+ 'compiler/control-flow-optimizer.h',
+ 'compiler/dead-code-elimination.cc',
+ 'compiler/dead-code-elimination.h',
+ 'compiler/diamond.h',
+ 'compiler/effect-control-linearizer.cc',
+ 'compiler/effect-control-linearizer.h',
+ 'compiler/escape-analysis.cc',
+ 'compiler/escape-analysis.h',
+ "compiler/escape-analysis-reducer.cc",
+ "compiler/escape-analysis-reducer.h",
+ 'compiler/frame.cc',
+ 'compiler/frame.h',
+ 'compiler/frame-elider.cc',
+ 'compiler/frame-elider.h',
+ "compiler/frame-states.cc",
+ "compiler/frame-states.h",
+ 'compiler/gap-resolver.cc',
+ 'compiler/gap-resolver.h',
+ 'compiler/graph-reducer.cc',
+ 'compiler/graph-reducer.h',
+ 'compiler/graph-replay.cc',
+ 'compiler/graph-replay.h',
+ 'compiler/graph-trimmer.cc',
+ 'compiler/graph-trimmer.h',
+ 'compiler/graph-visualizer.cc',
+ 'compiler/graph-visualizer.h',
+ 'compiler/graph.cc',
+ 'compiler/graph.h',
+ 'compiler/instruction-codes.h',
+ 'compiler/instruction-selector-impl.h',
+ 'compiler/instruction-selector.cc',
+ 'compiler/instruction-selector.h',
+ 'compiler/instruction-scheduler.cc',
+ 'compiler/instruction-scheduler.h',
+ 'compiler/instruction.cc',
+ 'compiler/instruction.h',
+ 'compiler/int64-lowering.cc',
+ 'compiler/int64-lowering.h',
+ 'compiler/js-builtin-reducer.cc',
+ 'compiler/js-builtin-reducer.h',
+ 'compiler/js-call-reducer.cc',
+ 'compiler/js-call-reducer.h',
+ 'compiler/js-context-specialization.cc',
+ 'compiler/js-context-specialization.h',
+ 'compiler/js-create-lowering.cc',
+ 'compiler/js-create-lowering.h',
+ 'compiler/js-frame-specialization.cc',
+ 'compiler/js-frame-specialization.h',
+ 'compiler/js-generic-lowering.cc',
+ 'compiler/js-generic-lowering.h',
+ 'compiler/js-global-object-specialization.cc',
+ 'compiler/js-global-object-specialization.h',
+ 'compiler/js-graph.cc',
+ 'compiler/js-graph.h',
+ 'compiler/js-inlining.cc',
+ 'compiler/js-inlining.h',
+ 'compiler/js-inlining-heuristic.cc',
+ 'compiler/js-inlining-heuristic.h',
+ 'compiler/js-intrinsic-lowering.cc',
+ 'compiler/js-intrinsic-lowering.h',
+ 'compiler/js-native-context-specialization.cc',
+ 'compiler/js-native-context-specialization.h',
+ 'compiler/js-operator.cc',
+ 'compiler/js-operator.h',
+ 'compiler/js-typed-lowering.cc',
+ 'compiler/js-typed-lowering.h',
+ 'compiler/jump-threading.cc',
+ 'compiler/jump-threading.h',
+ 'compiler/linkage.cc',
+ 'compiler/linkage.h',
+ 'compiler/liveness-analyzer.cc',
+ 'compiler/liveness-analyzer.h',
+ 'compiler/live-range-separator.cc',
+ 'compiler/live-range-separator.h',
+ 'compiler/load-elimination.cc',
+ 'compiler/load-elimination.h',
+ 'compiler/loop-analysis.cc',
+ 'compiler/loop-analysis.h',
+ 'compiler/loop-peeling.cc',
+ 'compiler/loop-peeling.h',
+ 'compiler/machine-operator-reducer.cc',
+ 'compiler/machine-operator-reducer.h',
+ 'compiler/machine-operator.cc',
+ 'compiler/machine-operator.h',
+ 'compiler/memory-optimizer.cc',
+ 'compiler/memory-optimizer.h',
+ 'compiler/move-optimizer.cc',
+ 'compiler/move-optimizer.h',
+ 'compiler/node-aux-data.h',
+ 'compiler/node-cache.cc',
+ 'compiler/node-cache.h',
+ 'compiler/node-marker.cc',
+ 'compiler/node-marker.h',
+ 'compiler/node-matchers.cc',
+ 'compiler/node-matchers.h',
+ 'compiler/node-properties.cc',
+ 'compiler/node-properties.h',
+ 'compiler/node.cc',
+ 'compiler/node.h',
+ 'compiler/opcodes.cc',
+ 'compiler/opcodes.h',
+ 'compiler/operation-typer.cc',
+ 'compiler/operation-typer.h',
+ 'compiler/operator-properties.cc',
+ 'compiler/operator-properties.h',
+ 'compiler/operator.cc',
+ 'compiler/operator.h',
+ 'compiler/osr.cc',
+ 'compiler/osr.h',
+ 'compiler/pipeline.cc',
+ 'compiler/pipeline.h',
+ 'compiler/pipeline-statistics.cc',
+ 'compiler/pipeline-statistics.h',
+ 'compiler/raw-machine-assembler.cc',
+ 'compiler/raw-machine-assembler.h',
+ 'compiler/redundancy-elimination.cc',
+ 'compiler/redundancy-elimination.h',
+ 'compiler/register-allocator.cc',
+ 'compiler/register-allocator.h',
+ 'compiler/register-allocator-verifier.cc',
+ 'compiler/register-allocator-verifier.h',
+ 'compiler/representation-change.cc',
+ 'compiler/representation-change.h',
+ 'compiler/schedule.cc',
+ 'compiler/schedule.h',
+ 'compiler/scheduler.cc',
+ 'compiler/scheduler.h',
+ 'compiler/select-lowering.cc',
+ 'compiler/select-lowering.h',
+ 'compiler/simplified-lowering.cc',
+ 'compiler/simplified-lowering.h',
+ 'compiler/simplified-operator-reducer.cc',
+ 'compiler/simplified-operator-reducer.h',
+ 'compiler/simplified-operator.cc',
+ 'compiler/simplified-operator.h',
+ 'compiler/source-position.cc',
+ 'compiler/source-position.h',
+ 'compiler/state-values-utils.cc',
+ 'compiler/state-values-utils.h',
+ 'compiler/store-store-elimination.cc',
+ 'compiler/store-store-elimination.h',
+ 'compiler/tail-call-optimization.cc',
+ 'compiler/tail-call-optimization.h',
+ 'compiler/type-hint-analyzer.cc',
+ 'compiler/type-hint-analyzer.h',
+ 'compiler/type-hints.cc',
+ 'compiler/type-hints.h',
+ 'compiler/typer.cc',
+ 'compiler/typer.h',
+ 'compiler/value-numbering-reducer.cc',
+ 'compiler/value-numbering-reducer.h',
+ 'compiler/verifier.cc',
+ 'compiler/verifier.h',
+ 'compiler/wasm-compiler.cc',
+ 'compiler/wasm-compiler.h',
+ 'compiler/wasm-linkage.cc',
+ 'compiler/zone-pool.cc',
+ 'compiler/zone-pool.h',
+ 'compiler.cc',
+ 'compiler.h',
+ 'context-measure.cc',
+ 'context-measure.h',
+ 'contexts-inl.h',
+ 'contexts.cc',
+ 'contexts.h',
+ 'conversions-inl.h',
+ 'conversions.cc',
+ 'conversions.h',
+ 'counters-inl.h',
+ 'counters.cc',
+ 'counters.h',
+ 'crankshaft/compilation-phase.cc',
+ 'crankshaft/compilation-phase.h',
+ 'crankshaft/hydrogen-alias-analysis.h',
+ 'crankshaft/hydrogen-bce.cc',
+ 'crankshaft/hydrogen-bce.h',
+ 'crankshaft/hydrogen-canonicalize.cc',
+ 'crankshaft/hydrogen-canonicalize.h',
+ 'crankshaft/hydrogen-check-elimination.cc',
+ 'crankshaft/hydrogen-check-elimination.h',
+ 'crankshaft/hydrogen-dce.cc',
+ 'crankshaft/hydrogen-dce.h',
+ 'crankshaft/hydrogen-dehoist.cc',
+ 'crankshaft/hydrogen-dehoist.h',
+ 'crankshaft/hydrogen-environment-liveness.cc',
+ 'crankshaft/hydrogen-environment-liveness.h',
+ 'crankshaft/hydrogen-escape-analysis.cc',
+ 'crankshaft/hydrogen-escape-analysis.h',
+ 'crankshaft/hydrogen-flow-engine.h',
+ 'crankshaft/hydrogen-gvn.cc',
+ 'crankshaft/hydrogen-gvn.h',
+ 'crankshaft/hydrogen-infer-representation.cc',
+ 'crankshaft/hydrogen-infer-representation.h',
+ 'crankshaft/hydrogen-infer-types.cc',
+ 'crankshaft/hydrogen-infer-types.h',
+ 'crankshaft/hydrogen-instructions.cc',
+ 'crankshaft/hydrogen-instructions.h',
+ 'crankshaft/hydrogen-load-elimination.cc',
+ 'crankshaft/hydrogen-load-elimination.h',
+ 'crankshaft/hydrogen-mark-deoptimize.cc',
+ 'crankshaft/hydrogen-mark-deoptimize.h',
+ 'crankshaft/hydrogen-mark-unreachable.cc',
+ 'crankshaft/hydrogen-mark-unreachable.h',
+ 'crankshaft/hydrogen-osr.cc',
+ 'crankshaft/hydrogen-osr.h',
+ 'crankshaft/hydrogen-range-analysis.cc',
+ 'crankshaft/hydrogen-range-analysis.h',
+ 'crankshaft/hydrogen-redundant-phi.cc',
+ 'crankshaft/hydrogen-redundant-phi.h',
+ 'crankshaft/hydrogen-removable-simulates.cc',
+ 'crankshaft/hydrogen-removable-simulates.h',
+ 'crankshaft/hydrogen-representation-changes.cc',
+ 'crankshaft/hydrogen-representation-changes.h',
+ 'crankshaft/hydrogen-sce.cc',
+ 'crankshaft/hydrogen-sce.h',
+ 'crankshaft/hydrogen-store-elimination.cc',
+ 'crankshaft/hydrogen-store-elimination.h',
+ 'crankshaft/hydrogen-types.cc',
+ 'crankshaft/hydrogen-types.h',
+ 'crankshaft/hydrogen-uint32-analysis.cc',
+ 'crankshaft/hydrogen-uint32-analysis.h',
+ 'crankshaft/hydrogen.cc',
+ 'crankshaft/hydrogen.h',
+ 'crankshaft/lithium-allocator-inl.h',
+ 'crankshaft/lithium-allocator.cc',
+ 'crankshaft/lithium-allocator.h',
+ 'crankshaft/lithium-codegen.cc',
+ 'crankshaft/lithium-codegen.h',
+ 'crankshaft/lithium.cc',
+ 'crankshaft/lithium.h',
+ 'crankshaft/lithium-inl.h',
+ 'crankshaft/typing.cc',
+ 'crankshaft/typing.h',
+ 'crankshaft/unique.h',
+ 'date.cc',
+ 'date.h',
+ 'dateparser-inl.h',
+ 'dateparser.cc',
+ 'dateparser.h',
+ 'debug/debug-evaluate.cc',
+ 'debug/debug-evaluate.h',
+ 'debug/debug-frames.cc',
+ 'debug/debug-frames.h',
+ 'debug/debug-scopes.cc',
+ 'debug/debug-scopes.h',
+ 'debug/debug.cc',
+ 'debug/debug.h',
+ 'debug/liveedit.cc',
+ 'debug/liveedit.h',
+ 'deoptimizer.cc',
+ 'deoptimizer.h',
+ 'disasm.h',
+ 'disassembler.cc',
+ 'disassembler.h',
+ 'diy-fp.cc',
+ 'diy-fp.h',
+ 'double.h',
+ 'dtoa.cc',
+ 'dtoa.h',
+ 'effects.h',
+ 'eh-frame.cc',
+ 'eh-frame.h',
+ 'elements-kind.cc',
+ 'elements-kind.h',
+ 'elements.cc',
+ 'elements.h',
+ 'execution.cc',
+ 'execution.h',
+ 'extensions/externalize-string-extension.cc',
+ 'extensions/externalize-string-extension.h',
+ 'extensions/free-buffer-extension.cc',
+ 'extensions/free-buffer-extension.h',
+ 'extensions/gc-extension.cc',
+ 'extensions/gc-extension.h',
+ 'extensions/ignition-statistics-extension.cc',
+ 'extensions/ignition-statistics-extension.h',
+ 'extensions/statistics-extension.cc',
+ 'extensions/statistics-extension.h',
+ 'extensions/trigger-failure-extension.cc',
+ 'extensions/trigger-failure-extension.h',
+ 'external-reference-table.cc',
+ 'external-reference-table.h',
+ 'factory.cc',
+ 'factory.h',
+ 'fast-accessor-assembler.cc',
+ 'fast-accessor-assembler.h',
+ 'fast-dtoa.cc',
+ 'fast-dtoa.h',
+ 'field-index.h',
+ 'field-index-inl.h',
+ 'field-type.cc',
+ 'field-type.h',
+ 'fixed-dtoa.cc',
+ 'fixed-dtoa.h',
+ 'flag-definitions.h',
+ 'flags.cc',
+ 'flags.h',
+ 'frames-inl.h',
+ 'frames.cc',
+ 'frames.h',
+ 'full-codegen/full-codegen.cc',
+ 'full-codegen/full-codegen.h',
+ 'futex-emulation.cc',
+ 'futex-emulation.h',
+ 'gdb-jit.cc',
+ 'gdb-jit.h',
+ 'global-handles.cc',
+ 'global-handles.h',
+ 'globals.h',
+ 'handles-inl.h',
+ 'handles.cc',
+ 'handles.h',
+ 'heap-symbols.h',
+ 'heap/array-buffer-tracker-inl.h',
+ 'heap/array-buffer-tracker.cc',
+ 'heap/array-buffer-tracker.h',
+ 'heap/memory-reducer.cc',
+ 'heap/memory-reducer.h',
+ 'heap/gc-idle-time-handler.cc',
+ 'heap/gc-idle-time-handler.h',
+ 'heap/gc-tracer.cc',
+ 'heap/gc-tracer.h',
+ 'heap/heap-inl.h',
+ 'heap/heap.cc',
+ 'heap/heap.h',
+ 'heap/incremental-marking-inl.h',
+ 'heap/incremental-marking-job.cc',
+ 'heap/incremental-marking-job.h',
+ 'heap/incremental-marking.cc',
+ 'heap/incremental-marking.h',
+ 'heap/mark-compact-inl.h',
+ 'heap/mark-compact.cc',
+ 'heap/mark-compact.h',
+ 'heap/object-stats.cc',
+ 'heap/object-stats.h',
+ 'heap/objects-visiting-inl.h',
+ 'heap/objects-visiting.cc',
+ 'heap/objects-visiting.h',
+ 'heap/page-parallel-job.h',
+ 'heap/remembered-set.cc',
+ 'heap/remembered-set.h',
+ 'heap/scavenge-job.h',
+ 'heap/scavenge-job.cc',
+ 'heap/scavenger-inl.h',
+ 'heap/scavenger.cc',
+ 'heap/scavenger.h',
+ 'heap/slot-set.h',
+ 'heap/spaces-inl.h',
+ 'heap/spaces.cc',
+ 'heap/spaces.h',
+ 'heap/store-buffer.cc',
+ 'heap/store-buffer.h',
+ 'i18n.cc',
+ 'i18n.h',
+ 'icu_util.cc',
+ 'icu_util.h',
+ 'ic/access-compiler.cc',
+ 'ic/access-compiler.h',
+ 'ic/call-optimization.cc',
+ 'ic/call-optimization.h',
+ 'ic/handler-compiler.cc',
+ 'ic/handler-compiler.h',
+ 'ic/ic-inl.h',
+ 'ic/ic-state.cc',
+ 'ic/ic-state.h',
+ 'ic/ic.cc',
+ 'ic/ic.h',
+ 'ic/ic-compiler.cc',
+ 'ic/ic-compiler.h',
+ 'identity-map.cc',
+ 'identity-map.h',
+ 'interface-descriptors.cc',
+ 'interface-descriptors.h',
+ 'interpreter/bytecodes.cc',
+ 'interpreter/bytecodes.h',
+ 'interpreter/bytecode-array-builder.cc',
+ 'interpreter/bytecode-array-builder.h',
+ 'interpreter/bytecode-array-iterator.cc',
+ 'interpreter/bytecode-array-iterator.h',
+ 'interpreter/bytecode-array-writer.cc',
+ 'interpreter/bytecode-array-writer.h',
+ 'interpreter/bytecode-dead-code-optimizer.cc',
+ 'interpreter/bytecode-dead-code-optimizer.h',
+ 'interpreter/bytecode-label.h',
+ 'interpreter/bytecode-generator.cc',
+ 'interpreter/bytecode-generator.h',
+ 'interpreter/bytecode-peephole-optimizer.cc',
+ 'interpreter/bytecode-peephole-optimizer.h',
+ 'interpreter/bytecode-pipeline.cc',
+ 'interpreter/bytecode-pipeline.h',
+ 'interpreter/bytecode-register-allocator.cc',
+ 'interpreter/bytecode-register-allocator.h',
+ 'interpreter/bytecode-register-optimizer.cc',
+ 'interpreter/bytecode-register-optimizer.h',
+ 'interpreter/bytecode-traits.h',
+ 'interpreter/constant-array-builder.cc',
+ 'interpreter/constant-array-builder.h',
+ 'interpreter/control-flow-builders.cc',
+ 'interpreter/control-flow-builders.h',
+ 'interpreter/handler-table-builder.cc',
+ 'interpreter/handler-table-builder.h',
+ 'interpreter/interpreter.cc',
+ 'interpreter/interpreter.h',
+ 'interpreter/interpreter-assembler.cc',
+ 'interpreter/interpreter-assembler.h',
+ 'interpreter/interpreter-intrinsics.cc',
+ 'interpreter/interpreter-intrinsics.h',
+ 'interpreter/source-position-table.cc',
+ 'interpreter/source-position-table.h',
+ 'isolate-inl.h',
+ 'isolate.cc',
+ 'isolate.h',
+ 'json-parser.cc',
+ 'json-parser.h',
+ 'json-stringifier.cc',
+ 'json-stringifier.h',
+ 'keys.h',
+ 'keys.cc',
+ 'layout-descriptor-inl.h',
+ 'layout-descriptor.cc',
+ 'layout-descriptor.h',
+ 'list-inl.h',
+ 'list.h',
+ 'locked-queue-inl.h',
+ 'locked-queue.h',
+ 'log-inl.h',
+ 'log-utils.cc',
+ 'log-utils.h',
+ 'log.cc',
+ 'log.h',
+ 'lookup.cc',
+ 'lookup.h',
+ 'macro-assembler.h',
+ 'machine-type.cc',
+ 'machine-type.h',
+ 'messages.cc',
+ 'messages.h',
+ 'msan.h',
+ 'objects-body-descriptors-inl.h',
+ 'objects-body-descriptors.h',
+ 'objects-debug.cc',
+ 'objects-inl.h',
+ 'objects-printer.cc',
+ 'objects.cc',
+ 'objects.h',
+ 'optimizing-compile-dispatcher.cc',
+ 'optimizing-compile-dispatcher.h',
+ 'ostreams.cc',
+ 'ostreams.h',
+ 'parsing/expression-classifier.h',
+ 'parsing/func-name-inferrer.cc',
+ 'parsing/func-name-inferrer.h',
+ 'parsing/parameter-initializer-rewriter.cc',
+ 'parsing/parameter-initializer-rewriter.h',
+ 'parsing/parser-base.h',
+ 'parsing/parser.cc',
+ 'parsing/parser.h',
+ 'parsing/pattern-rewriter.cc',
+ 'parsing/preparse-data-format.h',
+ 'parsing/preparse-data.cc',
+ 'parsing/preparse-data.h',
+ 'parsing/preparser.cc',
+ 'parsing/preparser.h',
+ 'parsing/rewriter.cc',
+ 'parsing/rewriter.h',
+ 'parsing/scanner-character-streams.cc',
+ 'parsing/scanner-character-streams.h',
+ 'parsing/scanner.cc',
+ 'parsing/scanner.h',
+ 'parsing/token.cc',
+ 'parsing/token.h',
+ 'pending-compilation-error-handler.cc',
+ 'pending-compilation-error-handler.h',
+ 'perf-jit.cc',
+ 'perf-jit.h',
+ 'profiler/allocation-tracker.cc',
+ 'profiler/allocation-tracker.h',
+ 'profiler/circular-queue-inl.h',
+ 'profiler/circular-queue.h',
+ 'profiler/cpu-profiler-inl.h',
+ 'profiler/cpu-profiler.cc',
+ 'profiler/cpu-profiler.h',
+ 'profiler/heap-profiler.cc',
+ 'profiler/heap-profiler.h',
+ 'profiler/heap-snapshot-generator-inl.h',
+ 'profiler/heap-snapshot-generator.cc',
+ 'profiler/heap-snapshot-generator.h',
+ 'profiler/profiler-listener.cc',
+ 'profiler/profiler-listener.h',
+ 'profiler/profile-generator-inl.h',
+ 'profiler/profile-generator.cc',
+ 'profiler/profile-generator.h',
+ 'profiler/sampling-heap-profiler.cc',
+ 'profiler/sampling-heap-profiler.h',
+ 'profiler/strings-storage.cc',
+ 'profiler/strings-storage.h',
+ 'profiler/tick-sample.cc',
+ 'profiler/tick-sample.h',
+ 'profiler/unbound-queue-inl.h',
+ 'profiler/unbound-queue.h',
+ 'property-descriptor.cc',
+ 'property-descriptor.h',
+ 'property-details.h',
+ 'property.cc',
+ 'property.h',
+ 'prototype.h',
+ 'regexp/bytecodes-irregexp.h',
+ 'regexp/interpreter-irregexp.cc',
+ 'regexp/interpreter-irregexp.h',
+ 'regexp/jsregexp-inl.h',
+ 'regexp/jsregexp.cc',
+ 'regexp/jsregexp.h',
+ 'regexp/regexp-ast.cc',
+ 'regexp/regexp-ast.h',
+ 'regexp/regexp-macro-assembler-irregexp-inl.h',
+ 'regexp/regexp-macro-assembler-irregexp.cc',
+ 'regexp/regexp-macro-assembler-irregexp.h',
+ 'regexp/regexp-macro-assembler-tracer.cc',
+ 'regexp/regexp-macro-assembler-tracer.h',
+ 'regexp/regexp-macro-assembler.cc',
+ 'regexp/regexp-macro-assembler.h',
+ 'regexp/regexp-parser.cc',
+ 'regexp/regexp-parser.h',
+ 'regexp/regexp-stack.cc',
+ 'regexp/regexp-stack.h',
+ 'register-configuration.cc',
+ 'register-configuration.h',
+ 'runtime-profiler.cc',
+ 'runtime-profiler.h',
+ 'runtime/runtime-array.cc',
+ 'runtime/runtime-atomics.cc',
+ 'runtime/runtime-classes.cc',
+ 'runtime/runtime-collections.cc',
+ 'runtime/runtime-compiler.cc',
+ 'runtime/runtime-date.cc',
+ 'runtime/runtime-debug.cc',
+ 'runtime/runtime-forin.cc',
+ 'runtime/runtime-function.cc',
+ 'runtime/runtime-futex.cc',
+ 'runtime/runtime-generator.cc',
+ 'runtime/runtime-i18n.cc',
+ 'runtime/runtime-internal.cc',
+ 'runtime/runtime-interpreter.cc',
+ 'runtime/runtime-literals.cc',
+ 'runtime/runtime-liveedit.cc',
+ 'runtime/runtime-maths.cc',
+ 'runtime/runtime-numbers.cc',
+ 'runtime/runtime-object.cc',
+ 'runtime/runtime-operators.cc',
+ 'runtime/runtime-proxy.cc',
+ 'runtime/runtime-regexp.cc',
+ 'runtime/runtime-scopes.cc',
+ 'runtime/runtime-simd.cc',
+ 'runtime/runtime-strings.cc',
+ 'runtime/runtime-symbol.cc',
+ 'runtime/runtime-test.cc',
+ 'runtime/runtime-typedarray.cc',
+ 'runtime/runtime-utils.h',
+ 'runtime/runtime.cc',
+ 'runtime/runtime.h',
+ 'safepoint-table.cc',
+ 'safepoint-table.h',
+ 'signature.h',
+ 'simulator.h',
+ 'small-pointer-list.h',
+ 'snapshot/code-serializer.cc',
+ 'snapshot/code-serializer.h',
+ 'snapshot/deserializer.cc',
+ 'snapshot/deserializer.h',
+ 'snapshot/natives.h',
+ 'snapshot/natives-common.cc',
+ 'snapshot/partial-serializer.cc',
+ 'snapshot/partial-serializer.h',
+ 'snapshot/serializer.cc',
+ 'snapshot/serializer.h',
+ 'snapshot/serializer-common.cc',
+ 'snapshot/serializer-common.h',
+ 'snapshot/snapshot.h',
+ 'snapshot/snapshot-common.cc',
+ 'snapshot/snapshot-source-sink.cc',
+ 'snapshot/snapshot-source-sink.h',
+ 'snapshot/startup-serializer.cc',
+ 'snapshot/startup-serializer.h',
+ 'source-position.h',
+ 'splay-tree.h',
+ 'splay-tree-inl.h',
+ 'startup-data-util.cc',
+ 'startup-data-util.h',
+ 'string-builder.cc',
+ 'string-builder.h',
+ 'string-search.h',
+ 'string-stream.cc',
+ 'string-stream.h',
+ 'strtod.cc',
+ 'strtod.h',
+ 'ic/stub-cache.cc',
+ 'ic/stub-cache.h',
+ 'tracing/trace-event.cc',
+ 'tracing/trace-event.h',
+ 'transitions-inl.h',
+ 'transitions.cc',
+ 'transitions.h',
+ 'type-cache.cc',
+ 'type-cache.h',
+ 'type-feedback-vector-inl.h',
+ 'type-feedback-vector.cc',
+ 'type-feedback-vector.h',
+ 'type-info.cc',
+ 'type-info.h',
+ 'types.cc',
+ 'types.h',
+ 'typing-asm.cc',
+ 'typing-asm.h',
+ 'unicode-inl.h',
+ 'unicode.cc',
+ 'unicode.h',
+ 'unicode-cache-inl.h',
+ 'unicode-cache.h',
+ 'unicode-decoder.cc',
+ 'unicode-decoder.h',
+ 'uri.cc',
+ 'uri.h',
+ 'utils-inl.h',
+ 'utils.cc',
+ 'utils.h',
+ 'v8.cc',
+ 'v8.h',
+ 'v8memory.h',
+ 'v8threads.cc',
+ 'v8threads.h',
+ 'vector.h',
+ 'version.cc',
+ 'version.h',
+ 'vm-state-inl.h',
+ 'vm-state.h',
+ 'wasm/asm-types.cc',
+ 'wasm/asm-types.h',
+ 'wasm/asm-wasm-builder.cc',
+ 'wasm/asm-wasm-builder.h',
+ 'wasm/ast-decoder.cc',
+ 'wasm/ast-decoder.h',
+ 'wasm/decoder.h',
+ 'wasm/encoder.cc',
+ 'wasm/encoder.h',
+ 'wasm/leb-helper.h',
+ 'wasm/module-decoder.cc',
+ 'wasm/module-decoder.h',
+ 'wasm/switch-logic.h',
+ 'wasm/switch-logic.cc',
+ 'wasm/wasm-debug.cc',
+ 'wasm/wasm-debug.h',
+ 'wasm/wasm-external-refs.cc',
+ 'wasm/wasm-external-refs.h',
+ 'wasm/wasm-function-name-table.cc',
+ 'wasm/wasm-function-name-table.h',
+ 'wasm/wasm-js.cc',
+ 'wasm/wasm-js.h',
+ 'wasm/wasm-macro-gen.h',
+ 'wasm/wasm-module.cc',
+ 'wasm/wasm-module.h',
+ 'wasm/wasm-interpreter.cc',
+ 'wasm/wasm-interpreter.h',
+ 'wasm/wasm-opcodes.cc',
+ 'wasm/wasm-opcodes.h',
+ 'wasm/wasm-result.cc',
+ 'wasm/wasm-result.h',
+ 'zone.cc',
+ 'zone.h',
+ 'zone-allocator.h',
+ 'zone-containers.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'conditions': [
+ ['OS!="win" and remove_v8base_debug_symbols==1', {
+ 'cflags!': ['-g', '-gdwarf-4'],
+ }],
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host', 'target'],
+ }, {
+ 'toolsets': ['target'],
+ }],
+ ['v8_target_arch=="arm"', {
+ 'sources': [ ### gcmole(arch:arm) ###
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'arm/assembler-arm-inl.h',
+ 'arm/assembler-arm.cc',
+ 'arm/assembler-arm.h',
+ 'arm/builtins-arm.cc',
+ 'arm/code-stubs-arm.cc',
+ 'arm/code-stubs-arm.h',
+ 'arm/codegen-arm.cc',
+ 'arm/codegen-arm.h',
+ 'arm/constants-arm.h',
+ 'arm/constants-arm.cc',
+ 'arm/cpu-arm.cc',
+ 'arm/deoptimizer-arm.cc',
+ 'arm/disasm-arm.cc',
+ 'arm/frames-arm.cc',
+ 'arm/frames-arm.h',
+ 'arm/interface-descriptors-arm.cc',
+ 'arm/interface-descriptors-arm.h',
+ 'arm/macro-assembler-arm.cc',
+ 'arm/macro-assembler-arm.h',
+ 'arm/simulator-arm.cc',
+ 'arm/simulator-arm.h',
+ 'compiler/arm/code-generator-arm.cc',
+ 'compiler/arm/instruction-codes-arm.h',
+ 'compiler/arm/instruction-scheduler-arm.cc',
+ 'compiler/arm/instruction-selector-arm.cc',
+ 'crankshaft/arm/lithium-arm.cc',
+ 'crankshaft/arm/lithium-arm.h',
+ 'crankshaft/arm/lithium-codegen-arm.cc',
+ 'crankshaft/arm/lithium-codegen-arm.h',
+ 'crankshaft/arm/lithium-gap-resolver-arm.cc',
+ 'crankshaft/arm/lithium-gap-resolver-arm.h',
+ 'debug/arm/debug-arm.cc',
+ 'full-codegen/arm/full-codegen-arm.cc',
+ 'ic/arm/access-compiler-arm.cc',
+ 'ic/arm/handler-compiler-arm.cc',
+ 'ic/arm/ic-arm.cc',
+ 'ic/arm/ic-compiler-arm.cc',
+ 'ic/arm/stub-cache-arm.cc',
+ 'regexp/arm/regexp-macro-assembler-arm.cc',
+ 'regexp/arm/regexp-macro-assembler-arm.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['v8_target_arch=="arm64"', {
+ 'sources': [ ### gcmole(arch:arm64) ###
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'arm64/assembler-arm64.cc',
+ 'arm64/assembler-arm64.h',
+ 'arm64/assembler-arm64-inl.h',
+ 'arm64/builtins-arm64.cc',
+ 'arm64/codegen-arm64.cc',
+ 'arm64/codegen-arm64.h',
+ 'arm64/code-stubs-arm64.cc',
+ 'arm64/code-stubs-arm64.h',
+ 'arm64/constants-arm64.h',
+ 'arm64/cpu-arm64.cc',
+ 'arm64/decoder-arm64.cc',
+ 'arm64/decoder-arm64.h',
+ 'arm64/decoder-arm64-inl.h',
+ 'arm64/deoptimizer-arm64.cc',
+ 'arm64/disasm-arm64.cc',
+ 'arm64/disasm-arm64.h',
+ 'arm64/frames-arm64.cc',
+ 'arm64/frames-arm64.h',
+ 'arm64/instructions-arm64.cc',
+ 'arm64/instructions-arm64.h',
+ 'arm64/instrument-arm64.cc',
+ 'arm64/instrument-arm64.h',
+ 'arm64/interface-descriptors-arm64.cc',
+ 'arm64/interface-descriptors-arm64.h',
+ 'arm64/macro-assembler-arm64.cc',
+ 'arm64/macro-assembler-arm64.h',
+ 'arm64/macro-assembler-arm64-inl.h',
+ 'arm64/simulator-arm64.cc',
+ 'arm64/simulator-arm64.h',
+ 'arm64/utils-arm64.cc',
+ 'arm64/utils-arm64.h',
+ 'compiler/arm64/code-generator-arm64.cc',
+ 'compiler/arm64/instruction-codes-arm64.h',
+ 'compiler/arm64/instruction-scheduler-arm64.cc',
+ 'compiler/arm64/instruction-selector-arm64.cc',
+ 'crankshaft/arm64/delayed-masm-arm64.cc',
+ 'crankshaft/arm64/delayed-masm-arm64.h',
+ 'crankshaft/arm64/delayed-masm-arm64-inl.h',
+ 'crankshaft/arm64/lithium-arm64.cc',
+ 'crankshaft/arm64/lithium-arm64.h',
+ 'crankshaft/arm64/lithium-codegen-arm64.cc',
+ 'crankshaft/arm64/lithium-codegen-arm64.h',
+ 'crankshaft/arm64/lithium-gap-resolver-arm64.cc',
+ 'crankshaft/arm64/lithium-gap-resolver-arm64.h',
+ 'debug/arm64/debug-arm64.cc',
+ 'full-codegen/arm64/full-codegen-arm64.cc',
+ 'ic/arm64/access-compiler-arm64.cc',
+ 'ic/arm64/handler-compiler-arm64.cc',
+ 'ic/arm64/ic-arm64.cc',
+ 'ic/arm64/ic-compiler-arm64.cc',
+ 'ic/arm64/stub-cache-arm64.cc',
+ 'regexp/arm64/regexp-macro-assembler-arm64.cc',
+ 'regexp/arm64/regexp-macro-assembler-arm64.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['v8_target_arch=="ia32"', {
+ 'sources': [ ### gcmole(arch:ia32) ###
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'ia32/assembler-ia32-inl.h',
+ 'ia32/assembler-ia32.cc',
+ 'ia32/assembler-ia32.h',
+ 'ia32/builtins-ia32.cc',
+ 'ia32/code-stubs-ia32.cc',
+ 'ia32/code-stubs-ia32.h',
+ 'ia32/codegen-ia32.cc',
+ 'ia32/codegen-ia32.h',
+ 'ia32/cpu-ia32.cc',
+ 'ia32/deoptimizer-ia32.cc',
+ 'ia32/disasm-ia32.cc',
+ 'ia32/frames-ia32.cc',
+ 'ia32/frames-ia32.h',
+ 'ia32/interface-descriptors-ia32.cc',
+ 'ia32/macro-assembler-ia32.cc',
+ 'ia32/macro-assembler-ia32.h',
+ 'compiler/ia32/code-generator-ia32.cc',
+ 'compiler/ia32/instruction-codes-ia32.h',
+ 'compiler/ia32/instruction-scheduler-ia32.cc',
+ 'compiler/ia32/instruction-selector-ia32.cc',
+ 'crankshaft/ia32/lithium-codegen-ia32.cc',
+ 'crankshaft/ia32/lithium-codegen-ia32.h',
+ 'crankshaft/ia32/lithium-gap-resolver-ia32.cc',
+ 'crankshaft/ia32/lithium-gap-resolver-ia32.h',
+ 'crankshaft/ia32/lithium-ia32.cc',
+ 'crankshaft/ia32/lithium-ia32.h',
+ 'debug/ia32/debug-ia32.cc',
+ 'full-codegen/ia32/full-codegen-ia32.cc',
+ 'ic/ia32/access-compiler-ia32.cc',
+ 'ic/ia32/handler-compiler-ia32.cc',
+ 'ic/ia32/ic-ia32.cc',
+ 'ic/ia32/ic-compiler-ia32.cc',
+ 'ic/ia32/stub-cache-ia32.cc',
+ 'regexp/ia32/regexp-macro-assembler-ia32.cc',
+ 'regexp/ia32/regexp-macro-assembler-ia32.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['v8_target_arch=="x87"', {
+ 'sources': [ ### gcmole(arch:x87) ###
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'x87/assembler-x87-inl.h',
+ 'x87/assembler-x87.cc',
+ 'x87/assembler-x87.h',
+ 'x87/builtins-x87.cc',
+ 'x87/code-stubs-x87.cc',
+ 'x87/code-stubs-x87.h',
+ 'x87/codegen-x87.cc',
+ 'x87/codegen-x87.h',
+ 'x87/cpu-x87.cc',
+ 'x87/deoptimizer-x87.cc',
+ 'x87/disasm-x87.cc',
+ 'x87/frames-x87.cc',
+ 'x87/frames-x87.h',
+ 'x87/interface-descriptors-x87.cc',
+ 'x87/macro-assembler-x87.cc',
+ 'x87/macro-assembler-x87.h',
+ 'compiler/x87/code-generator-x87.cc',
+ 'compiler/x87/instruction-codes-x87.h',
+ 'compiler/x87/instruction-scheduler-x87.cc',
+ 'compiler/x87/instruction-selector-x87.cc',
+ 'crankshaft/x87/lithium-codegen-x87.cc',
+ 'crankshaft/x87/lithium-codegen-x87.h',
+ 'crankshaft/x87/lithium-gap-resolver-x87.cc',
+ 'crankshaft/x87/lithium-gap-resolver-x87.h',
+ 'crankshaft/x87/lithium-x87.cc',
+ 'crankshaft/x87/lithium-x87.h',
+ 'debug/x87/debug-x87.cc',
+ 'full-codegen/x87/full-codegen-x87.cc',
+ 'ic/x87/access-compiler-x87.cc',
+ 'ic/x87/handler-compiler-x87.cc',
+ 'ic/x87/ic-x87.cc',
+ 'ic/x87/ic-compiler-x87.cc',
+ 'ic/x87/stub-cache-x87.cc',
+ 'regexp/x87/regexp-macro-assembler-x87.cc',
+ 'regexp/x87/regexp-macro-assembler-x87.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['v8_target_arch=="mips" or v8_target_arch=="mipsel"', {
+ 'sources': [ ### gcmole(arch:mipsel) ###
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'mips/assembler-mips.cc',
+ 'mips/assembler-mips.h',
+ 'mips/assembler-mips-inl.h',
+ 'mips/builtins-mips.cc',
+ 'mips/codegen-mips.cc',
+ 'mips/codegen-mips.h',
+ 'mips/code-stubs-mips.cc',
+ 'mips/code-stubs-mips.h',
+ 'mips/constants-mips.cc',
+ 'mips/constants-mips.h',
+ 'mips/cpu-mips.cc',
+ 'mips/deoptimizer-mips.cc',
+ 'mips/disasm-mips.cc',
+ 'mips/frames-mips.cc',
+ 'mips/frames-mips.h',
+ 'mips/interface-descriptors-mips.cc',
+ 'mips/macro-assembler-mips.cc',
+ 'mips/macro-assembler-mips.h',
+ 'mips/simulator-mips.cc',
+ 'mips/simulator-mips.h',
+ 'compiler/mips/code-generator-mips.cc',
+ 'compiler/mips/instruction-codes-mips.h',
+ 'compiler/mips/instruction-scheduler-mips.cc',
+ 'compiler/mips/instruction-selector-mips.cc',
+ 'crankshaft/mips/lithium-codegen-mips.cc',
+ 'crankshaft/mips/lithium-codegen-mips.h',
+ 'crankshaft/mips/lithium-gap-resolver-mips.cc',
+ 'crankshaft/mips/lithium-gap-resolver-mips.h',
+ 'crankshaft/mips/lithium-mips.cc',
+ 'crankshaft/mips/lithium-mips.h',
+ 'full-codegen/mips/full-codegen-mips.cc',
+ 'debug/mips/debug-mips.cc',
+ 'ic/mips/access-compiler-mips.cc',
+ 'ic/mips/handler-compiler-mips.cc',
+ 'ic/mips/ic-mips.cc',
+ 'ic/mips/ic-compiler-mips.cc',
+ 'ic/mips/stub-cache-mips.cc',
+ 'regexp/mips/regexp-macro-assembler-mips.cc',
+ 'regexp/mips/regexp-macro-assembler-mips.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ ['v8_target_arch=="mips64" or v8_target_arch=="mips64el"', {
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'sources': [ ### gcmole(arch:mips64el) ###
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'mips64/assembler-mips64.cc',
+ 'mips64/assembler-mips64.h',
+ 'mips64/assembler-mips64-inl.h',
+ 'mips64/builtins-mips64.cc',
+ 'mips64/codegen-mips64.cc',
+ 'mips64/codegen-mips64.h',
+ 'mips64/code-stubs-mips64.cc',
+ 'mips64/code-stubs-mips64.h',
+ 'mips64/constants-mips64.cc',
+ 'mips64/constants-mips64.h',
+ 'mips64/cpu-mips64.cc',
+ 'mips64/deoptimizer-mips64.cc',
+ 'mips64/disasm-mips64.cc',
+ 'mips64/frames-mips64.cc',
+ 'mips64/frames-mips64.h',
+ 'mips64/interface-descriptors-mips64.cc',
+ 'mips64/macro-assembler-mips64.cc',
+ 'mips64/macro-assembler-mips64.h',
+ 'mips64/simulator-mips64.cc',
+ 'mips64/simulator-mips64.h',
+ 'compiler/mips64/code-generator-mips64.cc',
+ 'compiler/mips64/instruction-codes-mips64.h',
+ 'compiler/mips64/instruction-scheduler-mips64.cc',
+ 'compiler/mips64/instruction-selector-mips64.cc',
+ 'crankshaft/mips64/lithium-codegen-mips64.cc',
+ 'crankshaft/mips64/lithium-codegen-mips64.h',
+ 'crankshaft/mips64/lithium-gap-resolver-mips64.cc',
+ 'crankshaft/mips64/lithium-gap-resolver-mips64.h',
+ 'crankshaft/mips64/lithium-mips64.cc',
+ 'crankshaft/mips64/lithium-mips64.h',
+ 'debug/mips64/debug-mips64.cc',
+ 'full-codegen/mips64/full-codegen-mips64.cc',
+ 'ic/mips64/access-compiler-mips64.cc',
+ 'ic/mips64/handler-compiler-mips64.cc',
+ 'ic/mips64/ic-mips64.cc',
+ 'ic/mips64/ic-compiler-mips64.cc',
+ 'ic/mips64/stub-cache-mips64.cc',
+ 'regexp/mips64/regexp-macro-assembler-mips64.cc',
+ 'regexp/mips64/regexp-macro-assembler-mips64.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['v8_target_arch=="x64" or v8_target_arch=="x32"', {
+ 'sources': [ ### gcmole(arch:x64) ###
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'crankshaft/x64/lithium-codegen-x64.cc',
+ 'crankshaft/x64/lithium-codegen-x64.h',
+ 'crankshaft/x64/lithium-gap-resolver-x64.cc',
+ 'crankshaft/x64/lithium-gap-resolver-x64.h',
+ 'crankshaft/x64/lithium-x64.cc',
+ 'crankshaft/x64/lithium-x64.h',
+ 'x64/assembler-x64-inl.h',
+ 'x64/assembler-x64.cc',
+ 'x64/assembler-x64.h',
+ 'x64/builtins-x64.cc',
+ 'x64/code-stubs-x64.cc',
+ 'x64/code-stubs-x64.h',
+ 'x64/codegen-x64.cc',
+ 'x64/codegen-x64.h',
+ 'x64/cpu-x64.cc',
+ 'x64/deoptimizer-x64.cc',
+ 'x64/disasm-x64.cc',
+ 'x64/frames-x64.cc',
+ 'x64/frames-x64.h',
+ 'x64/interface-descriptors-x64.cc',
+ 'x64/macro-assembler-x64.cc',
+ 'x64/macro-assembler-x64.h',
+ 'debug/x64/debug-x64.cc',
+ 'full-codegen/x64/full-codegen-x64.cc',
+ 'ic/x64/access-compiler-x64.cc',
+ 'ic/x64/handler-compiler-x64.cc',
+ 'ic/x64/ic-x64.cc',
+ 'ic/x64/ic-compiler-x64.cc',
+ 'ic/x64/stub-cache-x64.cc',
+ 'regexp/x64/regexp-macro-assembler-x64.cc',
+ 'regexp/x64/regexp-macro-assembler-x64.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['v8_target_arch=="x64"', {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'compiler/x64/code-generator-x64.cc',
+ 'compiler/x64/instruction-codes-x64.h',
+ 'compiler/x64/instruction-scheduler-x64.cc',
+ 'compiler/x64/instruction-selector-x64.cc',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['v8_target_arch=="ppc" or v8_target_arch=="ppc64"', {
+ 'sources': [ ### gcmole(arch:ppc) ###
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'compiler/ppc/code-generator-ppc.cc',
+ 'compiler/ppc/instruction-codes-ppc.h',
+ 'compiler/ppc/instruction-scheduler-ppc.cc',
+ 'compiler/ppc/instruction-selector-ppc.cc',
+ 'crankshaft/ppc/lithium-ppc.cc',
+ 'crankshaft/ppc/lithium-ppc.h',
+ 'crankshaft/ppc/lithium-codegen-ppc.cc',
+ 'crankshaft/ppc/lithium-codegen-ppc.h',
+ 'crankshaft/ppc/lithium-gap-resolver-ppc.cc',
+ 'crankshaft/ppc/lithium-gap-resolver-ppc.h',
+ 'debug/ppc/debug-ppc.cc',
+ 'full-codegen/ppc/full-codegen-ppc.cc',
+ 'ic/ppc/access-compiler-ppc.cc',
+ 'ic/ppc/handler-compiler-ppc.cc',
+ 'ic/ppc/ic-ppc.cc',
+ 'ic/ppc/ic-compiler-ppc.cc',
+ 'ic/ppc/stub-cache-ppc.cc',
+ 'ppc/assembler-ppc-inl.h',
+ 'ppc/assembler-ppc.cc',
+ 'ppc/assembler-ppc.h',
+ 'ppc/builtins-ppc.cc',
+ 'ppc/code-stubs-ppc.cc',
+ 'ppc/code-stubs-ppc.h',
+ 'ppc/codegen-ppc.cc',
+ 'ppc/codegen-ppc.h',
+ 'ppc/constants-ppc.h',
+ 'ppc/constants-ppc.cc',
+ 'ppc/cpu-ppc.cc',
+ 'ppc/deoptimizer-ppc.cc',
+ 'ppc/disasm-ppc.cc',
+ 'ppc/frames-ppc.cc',
+ 'ppc/frames-ppc.h',
+ 'ppc/interface-descriptors-ppc.cc',
+ 'ppc/macro-assembler-ppc.cc',
+ 'ppc/macro-assembler-ppc.h',
+ 'ppc/simulator-ppc.cc',
+ 'ppc/simulator-ppc.h',
+ 'regexp/ppc/regexp-macro-assembler-ppc.cc',
+ 'regexp/ppc/regexp-macro-assembler-ppc.h',
+ ],
+ }],
+ ['v8_target_arch=="s390" or v8_target_arch=="s390x"', {
+ 'sources': [ ### gcmole(arch:s390) ###
+ 'compiler/s390/code-generator-s390.cc',
+ 'compiler/s390/instruction-codes-s390.h',
+ 'compiler/s390/instruction-scheduler-s390.cc',
+ 'compiler/s390/instruction-selector-s390.cc',
+ 'crankshaft/s390/lithium-codegen-s390.cc',
+ 'crankshaft/s390/lithium-codegen-s390.h',
+ 'crankshaft/s390/lithium-gap-resolver-s390.cc',
+ 'crankshaft/s390/lithium-gap-resolver-s390.h',
+ 'crankshaft/s390/lithium-s390.cc',
+ 'crankshaft/s390/lithium-s390.h',
+ 'debug/s390/debug-s390.cc',
+ 'full-codegen/s390/full-codegen-s390.cc',
+ 'ic/s390/access-compiler-s390.cc',
+ 'ic/s390/handler-compiler-s390.cc',
+ 'ic/s390/ic-compiler-s390.cc',
+ 'ic/s390/ic-s390.cc',
+ 'ic/s390/stub-cache-s390.cc',
+ 'regexp/s390/regexp-macro-assembler-s390.cc',
+ 'regexp/s390/regexp-macro-assembler-s390.h',
+ 's390/assembler-s390.cc',
+ 's390/assembler-s390.h',
+ 's390/assembler-s390-inl.h',
+ 's390/builtins-s390.cc',
+ 's390/codegen-s390.cc',
+ 's390/codegen-s390.h',
+ 's390/code-stubs-s390.cc',
+ 's390/code-stubs-s390.h',
+ 's390/constants-s390.cc',
+ 's390/constants-s390.h',
+ 's390/cpu-s390.cc',
+ 's390/deoptimizer-s390.cc',
+ 's390/disasm-s390.cc',
+ 's390/frames-s390.cc',
+ 's390/frames-s390.h',
+ 's390/interface-descriptors-s390.cc',
+ 's390/macro-assembler-s390.cc',
+ 's390/macro-assembler-s390.h',
+ 's390/simulator-s390.cc',
+ 's390/simulator-s390.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['OS=="win"', {
+ 'variables': {
+ 'gyp_generators': '<!(echo $GYP_GENERATORS)',
+ },
+ 'msvs_disabled_warnings': [4351, 4355, 4800],
+ # When building Official, the .lib is too large and exceeds the 2G
+ # limit. This breaks it into multiple pieces to avoid the limit.
+ # See http://crbug.com/485155.
+ 'msvs_shard': 4,
+ }],
+ ['v8_component=="shared_library"', {
+ 'defines': [
+ 'BUILDING_V8_SHARED',
+ 'V8_SHARED',
+ ],
+ }],
+ ['v8_postmortem_support=="true"', {
+ 'sources': [
+ '<(SHARED_INTERMEDIATE_DIR)/debug-support.cc',
+ ]
+ }],
+ ['v8_enable_i18n_support==1', {
+ 'dependencies': [
+ '<(icu_gyp_path):icui18n',
+ '<(icu_gyp_path):icuuc',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ ],
+ 'conditions': [
+ ['icu_use_data_file_flag==1', {
+ 'defines': ['ICU_UTIL_DATA_IMPL=ICU_UTIL_DATA_FILE'],
+ }, { # else icu_use_data_file_flag !=1
+ 'conditions': [
+ ['OS=="win"', {
+ 'defines': ['ICU_UTIL_DATA_IMPL=ICU_UTIL_DATA_SHARED'],
+ }, {
+ 'defines': ['ICU_UTIL_DATA_IMPL=ICU_UTIL_DATA_STATIC'],
+ }],
+ ],
+ }],
+ ],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ }, { # v8_enable_i18n_support==0
+ 'sources!': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'i18n.cc',
+ 'i18n.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['OS=="win" and v8_enable_i18n_support==1', {
+ 'dependencies': [
+ '<(icu_gyp_path):icudata',
+ ],
+ }],
+ ],
+ },
+ {
+ 'target_name': 'v8_libbase_sse2',
+ 'type': 'static_library',
+ 'variables': {
+ 'optimize': 'max',
+ },
+ 'include_dirs+': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '..',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/accounting-allocator.cc',
+ 'base/accounting-allocator.h',
+ 'base/adapters.h',
+ 'base/atomic-utils.h',
+ 'base/atomicops.h',
+ 'base/atomicops_internals_arm64_gcc.h',
+ 'base/atomicops_internals_arm_gcc.h',
+ 'base/atomicops_internals_atomicword_compat.h',
+ 'base/atomicops_internals_mac.h',
+ 'base/atomicops_internals_mips_gcc.h',
+ 'base/atomicops_internals_mips64_gcc.h',
+ 'base/atomicops_internals_portable.h',
+ 'base/atomicops_internals_ppc_gcc.h',
+ 'base/atomicops_internals_s390_gcc.h',
+ 'base/atomicops_internals_tsan.h',
+ 'base/atomicops_internals_x86_gcc.cc',
+ 'base/atomicops_internals_x86_gcc.h',
+ 'base/atomicops_internals_x86_msvc.h',
+ 'base/bits.cc',
+ 'base/bits.h',
+ 'base/build_config.h',
+ 'base/compiler-specific.h',
+ 'base/cpu.cc',
+ 'base/cpu.h',
+ 'base/division-by-constant.cc',
+ 'base/division-by-constant.h',
+ 'base/file-utils.cc',
+ 'base/file-utils.h',
+ 'base/flags.h',
+ 'base/format-macros.h',
+ 'base/functional.cc',
+ 'base/functional.h',
+ 'base/hashmap.h',
+ 'base/ieee754.cc',
+ 'base/ieee754.h',
+ 'base/iterator.h',
+ 'base/lazy-instance.h',
+ 'base/logging.cc',
+ 'base/logging.h',
+ 'base/macros.h',
+ 'base/once.cc',
+ 'base/once.h',
+ 'base/platform/elapsed-timer.h',
+ 'base/platform/time.cc',
+ 'base/platform/time.h',
+ 'base/platform/condition-variable.cc',
+ 'base/platform/condition-variable.h',
+ 'base/platform/mutex.cc',
+ 'base/platform/mutex.h',
+ 'base/platform/platform.h',
+ 'base/platform/semaphore.cc',
+ 'base/platform/semaphore.h',
+ 'base/safe_conversions.h',
+ 'base/safe_conversions_impl.h',
+ 'base/safe_math.h',
+ 'base/safe_math_impl.h',
+ 'base/smart-pointers.h',
+ 'base/sys-info.cc',
+ 'base/sys-info.h',
+ 'base/utils/random-number-generator.cc',
+ 'base/utils/random-number-generator.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'conditions': [
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host', 'target'],
+ }, {
+ 'toolsets': ['target'],
+ }],
+ ['OS=="linux"', {
+ 'conditions': [
+ ['nacl_target_arch=="none"', {
+ 'link_settings': {
+ 'libraries': [
+ '-ldl',
+ '-lrt'
+ ],
+ },
+ }, {
+ 'defines': [
+ 'V8_LIBRT_NOT_AVAILABLE=1',
+ ],
+ }],
+ ],
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-linux.cc',
+ 'base/platform/platform-posix.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }
+ ],
+ ['OS=="android"', {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-posix.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'link_settings': {
+ 'target_conditions': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ ['_toolset=="host" and host_os!="mac"', {
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ # Only include libdl and librt on host builds because they
+ # are included by default on Android target builds, and we
+ # don't want to re-include them here since this will change
+ # library order and break (see crbug.com/469973).
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ # These libraries do not exist on Mac hosted builds.
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'libraries': [
+ '-ldl',
+ '-lrt'
+ ]
+ }]
+ ]
+ },
+ 'conditions': [
+ ['host_os=="mac"', {
+ 'target_conditions': [
+ ['_toolset=="host"', {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-macos.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ]
+ }, {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-linux.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ]
+ }],
+ ],
+ }, {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-linux.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ]
+ }],
+ ],
+ },
+ ],
+ ['OS=="qnx"', {
+ 'link_settings': {
+ 'target_conditions': [
+ ['_toolset=="host" and host_os=="linux"', {
+ 'libraries': [
+ '-lrt'
+ ],
+ }],
+ ['_toolset=="target"', {
+ 'libraries': [
+ '-lbacktrace'
+ ],
+ }],
+ ],
+ },
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-posix.cc',
+ 'base/qnx-math.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'target_conditions': [
+ ['_toolset=="host" and host_os=="linux"', {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-linux.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['_toolset=="host" and host_os=="mac"', {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-macos.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ['_toolset=="target"', {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-qnx.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ],
+ },
+ ],
+ ['OS=="freebsd"', {
+ 'link_settings': {
+ 'libraries': [
+ '-L/usr/local/lib -lexecinfo',
+ ]},
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-freebsd.cc',
+ 'base/platform/platform-posix.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }
+ ],
+ ['OS=="openbsd"', {
+ 'link_settings': {
+ 'libraries': [
+ '-L/usr/local/lib -lexecinfo',
+ ]},
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-openbsd.cc',
+ 'base/platform/platform-posix.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }
+ ],
+ ['OS=="netbsd"', {
+ 'link_settings': {
+ 'libraries': [
+ '-L/usr/pkg/lib -Wl,-R/usr/pkg/lib -lexecinfo',
+ ]},
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-openbsd.cc',
+ 'base/platform/platform-posix.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }
+ ],
+ ['OS=="aix"', {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-aix.cc',
+ 'base/platform/platform-posix.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ]},
+ ],
+ ['OS=="solaris"', {
+ 'link_settings': {
+ 'libraries': [
+ '-lnsl -lrt',
+ ]},
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-solaris.cc',
+ 'base/platform/platform-posix.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }
+ ],
+ ['OS=="mac"', {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-macos.cc',
+ 'base/platform/platform-posix.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ]},
+ ],
+ ['OS=="win"', {
+ 'defines': [
+ '_CRT_RAND_S' # for rand_s()
+ ],
+ 'variables': {
+ 'gyp_generators': '<!(echo $GYP_GENERATORS)',
+ },
+ 'conditions': [
+ ['gyp_generators=="make"', {
+ 'variables': {
+ 'build_env': '<!(uname -o)',
+ },
+ 'conditions': [
+ ['build_env=="Cygwin"', {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-cygwin.cc',
+ 'base/platform/platform-posix.cc'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }, {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-win32.cc',
+ 'base/win32-headers.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ],
+ 'link_settings': {
+ 'libraries': [ '-lwinmm', '-lws2_32' ],
+ },
+ }, {
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'base/platform/platform-win32.cc',
+ 'base/win32-headers.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'msvs_disabled_warnings': [4351, 4355, 4800],
+ 'link_settings': {
+ 'libraries': [ '-lwinmm.lib', '-lws2_32.lib' ],
+ },
+ }],
+ ],
+ }],
+ ],
+ },
+ {
+ 'target_name': 'v8_libplatform_sse2',
+ 'type': 'static_library',
+ 'variables': {
+ 'optimize': 'max',
+ },
+ 'dependencies': [
+ 'v8_libbase_sse2',
+ ],
+ 'include_dirs+': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '..',
+ '../include',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../include/libplatform/libplatform.h',
+ 'libplatform/default-platform.cc',
+ 'libplatform/default-platform.h',
+ 'libplatform/task-queue.cc',
+ 'libplatform/task-queue.h',
+ 'libplatform/worker-thread.cc',
+ 'libplatform/worker-thread.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'conditions': [
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host', 'target'],
+ }, {
+ 'toolsets': ['target'],
+ }],
+ ],
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'direct_dependent_settings': {
+ 'include_dirs': [
+ '../include',
+ ],
+ },
+ },
+ {
+ 'target_name': 'v8_libsampler_sse2',
+ 'type': 'static_library',
+ 'variables': {
+ 'optimize': 'max',
+ },
+ 'dependencies': [
+ 'v8_libbase_sse2',
+ ],
+ 'include_dirs+': [
+ '..',
+ '../include',
+ ],
+ 'sources': [
+ 'libsampler/v8-sampler.cc',
+ 'libsampler/v8-sampler.h'
+ ],
+ 'conditions': [
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host', 'target'],
+ }, {
+ 'toolsets': ['target'],
+ }],
+ ],
+ 'direct_dependent_settings': {
+ 'include_dirs': [
+ '../include',
+ ],
+ },
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ },
+ {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'target_name': 'natives_blob_sse2',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'type': 'none',
+ 'conditions': [
+ [ 'v8_use_external_startup_data==1', {
+ 'conditions': [
+ ['want_separate_host_toolset==1', {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'dependencies': ['js2c_sse2#host'],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ }, {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'dependencies': ['js2c_sse2'],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ }],
+ ],
+ 'actions': [{
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'action_name': 'v8_sse2_concatenate_natives_blob',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'inputs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/concatenate-files.py',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/libraries_sse2.bin',
+ '<(SHARED_INTERMEDIATE_DIR)/libraries-experimental_sse2.bin',
+ '<(SHARED_INTERMEDIATE_DIR)/libraries-extras_sse2.bin',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/libraries-experimental-extras_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'conditions': [
+ ['want_separate_host_toolset==1', {
+ 'target_conditions': [
+ ['_toolset=="host"', {
+ 'outputs': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(PRODUCT_DIR)/natives_blob_host_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'action': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'python', '<@(_inputs)', '<(PRODUCT_DIR)/natives_blob_host_sse2.bin'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }, {
+ 'outputs': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(PRODUCT_DIR)/natives_blob_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'action': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'python', '<@(_inputs)', '<(PRODUCT_DIR)/natives_blob_sse2.bin'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ],
+ }, {
+ 'outputs': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(PRODUCT_DIR)/natives_blob_sse2.bin',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'action': [
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'python', '<@(_inputs)', '<(PRODUCT_DIR)/natives_blob_sse2.bin'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ }],
+ ],
+ }],
+ }],
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host', 'target'],
+ }, {
+ 'toolsets': ['target'],
+ }],
+ ]
+ },
+ {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'target_name': 'js2c_sse2',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'type': 'none',
+ 'conditions': [
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host'],
+ }, {
+ 'toolsets': ['target'],
+ }],
+ ],
+ 'variables': {
+ 'library_files': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'js/macros.py',
+ 'messages.h',
+ 'js/prologue.js',
+ 'js/runtime.js',
+ 'js/v8natives.js',
+ 'js/symbol.js',
+ 'js/array.js',
+ 'js/string.js',
+ 'js/math.js',
+ 'third_party/fdlibm/fdlibm.js',
+ 'js/regexp.js',
+ 'js/arraybuffer.js',
+ 'js/typedarray.js',
+ 'js/iterator-prototype.js',
+ 'js/collection.js',
+ 'js/weak-collection.js',
+ 'js/collection-iterator.js',
+ 'js/promise.js',
+ 'js/messages.js',
+ 'js/array-iterator.js',
+ 'js/string-iterator.js',
+ 'js/templates.js',
+ 'js/spread.js',
+ 'js/proxy.js',
+ 'debug/mirrors.js',
+ 'debug/debug.js',
+ 'debug/liveedit.js',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'experimental_library_files': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'js/macros.py',
+ 'messages.h',
+ 'js/harmony-atomics.js',
+ 'js/harmony-sharedarraybuffer.js',
+ 'js/harmony-simd.js',
+ 'js/harmony-string-padding.js',
+ 'js/promise-extra.js',
+ 'js/harmony-async-await.js'
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'libraries_bin_file': '<(SHARED_INTERMEDIATE_DIR)/libraries_sse2.bin',
+ 'libraries_experimental_bin_file': '<(SHARED_INTERMEDIATE_DIR)/libraries-experimental_sse2.bin',
+ 'libraries_extras_bin_file': '<(SHARED_INTERMEDIATE_DIR)/libraries-extras_sse2.bin',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ 'libraries_experimental_extras_bin_file': '<(SHARED_INTERMEDIATE_DIR)/libraries-experimental-extras_sse2.bin',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'conditions': [
+ ['v8_enable_i18n_support==1', {
+ 'library_files': ['js/i18n.js'],
+ 'experimental_library_files': [
+ 'js/icu-case-mapping.js',
+ 'js/intl-extra.js',
+ ],
+ }],
+ ],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ },
+ 'actions': [
+ {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'action_name': 'v8_sse2_js2c',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'inputs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ '<@(library_files)',
+ ],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ 'outputs': ['<(SHARED_INTERMEDIATE_DIR)/libraries_sse2.cc'],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'action': [
+ 'python',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/libraries_sse2.cc',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'CORE',
+ '<@(library_files)',
+ ],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ },
+ {
+ 'action_name': 'v8_sse2_js2c_bin',
+ 'inputs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<@(library_files)',
+ ],
+ 'outputs': ['<@(libraries_bin_file)'],
+ 'action': [
+ 'python',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/libraries_sse2.cc',
+ 'CORE',
+ '<@(library_files)',
+ '--startup_blob', '<@(libraries_bin_file)',
+ '--nojs',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ },
+ {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'action_name': 'v8_sse2_js2c_experimental',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'inputs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ '<@(experimental_library_files)',
+ ],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ 'outputs': ['<(SHARED_INTERMEDIATE_DIR)/experimental-libraries_sse2.cc'],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'action': [
+ 'python',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/experimental-libraries_sse2.cc',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'EXPERIMENTAL',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '<@(experimental_library_files)',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ },
+ {
+ 'action_name': 'v8_sse2_js2c_experimental_bin',
+ 'inputs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<@(experimental_library_files)',
+ ],
+ 'outputs': ['<@(libraries_experimental_bin_file)'],
+ 'action': [
+ 'python',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/experimental-libraries_sse2.cc',
+ 'EXPERIMENTAL',
+ '<@(experimental_library_files)',
+ '--startup_blob', '<@(libraries_experimental_bin_file)',
+ '--nojs',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ },
+ {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'action_name': 'v8_sse2_js2c_extras',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'inputs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ '<@(v8_extra_library_files)',
+ ],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ 'outputs': ['<(SHARED_INTERMEDIATE_DIR)/extras-libraries_sse2.cc'],
+ 'action': [
+ 'python',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/extras-libraries_sse2.cc',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ 'EXTRAS',
+ '<@(v8_extra_library_files)',
+ ],
+ },
+ {
+ 'action_name': 'v8_sse2_js2c_extras_bin',
+ 'inputs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<@(v8_extra_library_files)',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ 'outputs': ['<@(libraries_extras_bin_file)'],
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'action': [
+ 'python',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/extras-libraries_sse2.cc',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'EXTRAS',
+ '<@(v8_extra_library_files)',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '--startup_blob', '<@(libraries_extras_bin_file)',
+ '--nojs',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ },
+ {
+ 'action_name': 'v8_sse2_js2c_experimental_extras',
+ 'inputs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<@(v8_experimental_extra_library_files)',
+ ],
+ 'outputs': [
+ '<(SHARED_INTERMEDIATE_DIR)/experimental-extras-libraries_sse2.cc',
+ ],
+ 'action': [
+ 'python',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/experimental-extras-libraries_sse2.cc',
+ 'EXPERIMENTAL_EXTRAS',
+ '<@(v8_experimental_extra_library_files)',
+ ],
+ },
+ {
+ 'action_name': 'v8_sse2_js2c_experimental_extras_bin',
+ 'inputs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<@(v8_experimental_extra_library_files)',
+ ],
+ 'outputs': ['<@(libraries_experimental_extras_bin_file)'],
+ 'action': [
+ 'python',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/js2c.py',
Rebase the no-sse2 patch Now all patches should be rebased.
2016-07-17 19:06:36 +00:00
+ '<(SHARED_INTERMEDIATE_DIR)/experimental-extras-libraries_sse2.cc',
+ 'EXPERIMENTAL_EXTRAS',
+ '<@(v8_experimental_extra_library_files)',
+ '--startup_blob', '<@(libraries_experimental_extras_bin_file)',
+ '--nojs',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ },
+ ],
+ },
+ {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'target_name': 'postmortem-metadata_sse2',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'type': 'none',
+ 'variables': {
+ 'heapobject_files': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'objects.h',
+ 'objects-inl.h',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ },
+ 'actions': [
+ {
no-sse2.patch: More v8_sse2.gyp fixes, ensure everything is renamed (again, by fixing the make-v8-sse2-gyp.sh script)
2016-01-19 03:58:50 +00:00
+ 'action_name': 'v8_sse2_gen-postmortem-metadata',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ 'inputs': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/gen-postmortem-metadata.py',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ '<@(heapobject_files)',
+ ],
+ 'outputs': [
+ '<(SHARED_INTERMEDIATE_DIR)/debug-support.cc',
+ ],
+ 'action': [
+ 'python',
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '../tools/gen-postmortem-metadata.py',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ '<@(_outputs)',
+ '<@(heapobject_files)'
+ ]
+ }
+ ]
+ },
+ {
+ 'target_name': 'mksnapshot_sse2',
+ 'type': 'executable',
+ 'dependencies': ['v8_base_sse2', 'v8_nosnapshot_sse2', 'v8_libplatform_sse2'],
+ 'include_dirs+': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ '..',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'sources': [
Update to 5.8.0 - Fix FTBFS in the WTF part of Blink/WebKit with GCC 7
2017-03-26 19:48:05 +00:00
+ 'snapshot/mksnapshot.cc',
no-sse2.patch: Build both the x87 version and the SSE2 version of V8 on i686 Note that almost all the patch size increase is from the v8_sse2.gyp file that is automatically generated from v8.gyp using a small script I wrote, which is also included in the patch.
2016-01-19 02:32:24 +00:00
+ ],
+ 'conditions': [
+ ['v8_enable_i18n_support==1', {
+ 'dependencies': [
+ '<(icu_gyp_path):icui18n',
+ '<(icu_gyp_path):icuuc',
+ ]
+ }],
+ ['want_separate_host_toolset==1', {
+ 'toolsets': ['host'],
+ }, {
+ 'toolsets': ['target'],
+ }],
+ ],
+ },
+ ],
+}
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
diff --git a/src/core/core_module.pro b/src/core/core_module.pro
index 44e8ac613..7fe1a49e9 100644
--- a/src/core/core_module.pro
+++ b/src/core/core_module.pro
@@ -10,6 +10,29 @@ linking_pri = $$OUT_PWD/$$getConfigDir()/$${TARGET}.pri
error("Could not find the linking information that gn should have generated.")
no-sse2.patch: Install the shared libraries, don't link the sse2 one
2016-01-19 13:55:13 +00:00
}
+# find the shared libraries in the link line
+# remove those in lib/sse2 that are only replacements for the normal ones
+# collect all shared libraries so they can be installed
+for(private_lib, LIBS_PRIVATE) {
+ contains(private_lib, .*\.so) {
+ contains(private_lib, .*/lib/sse2/.*) {
+ LIBS_PRIVATE -= $$private_lib
+ shlibs_sse2 += $$private_lib
+ } else {
+ shlibs += $$private_lib
+ }
+ }
+}
+
+# set the shared libraries to be installed
+# add an rpath to their installation location
+shlib_install_path = $$[QT_INSTALL_LIBS]/qtwebengine
+!isEmpty(shlibs) {
+ shlibs.files += $$shlibs
+ shlibs_sse2.files += $$shlibs_sse2
+ LIBS_PRIVATE += -Wl,--rpath,$$shlib_install_path
+}
+
load(qt_module)
api_library_name = qtwebenginecoreapi$$qtPlatformTargetSuffix()
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
@@ -100,7 +123,12 @@ icu.files = $$OUT_PWD/$$getConfigDir()/icudtl.dat
Update to 5.6.1 * Fri Jun 10 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.1-1 - Update to 5.6.1 - Rebase linux-pri patch (drop the parts already fixed upstream) - Drop backported chimera-nss-init patch, already applied upstream - Rebase no-sse2 patch (the core_module.pro change)
2016-06-10 23:32:18 +00:00
locales.path = $$[QT_INSTALL_TRANSLATIONS]/qtwebengine_locales
resources.CONFIG += no_check_exist
Rebase no-sse2 patch, fix file list - Rebase no-sse2 patch - Update file list for 5.6.0 RC (resources now in resources/ subdirectory) - Tag translations with correct %lang tags
2016-02-25 00:24:47 +00:00
resources.path = $$[QT_INSTALL_DATA]/resources
Update to 5.6.1 * Fri Jun 10 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.1-1 - Update to 5.6.1 - Rebase linux-pri patch (drop the parts already fixed upstream) - Drop backported chimera-nss-init patch, already applied upstream - Rebase no-sse2 patch (the core_module.pro change)
2016-06-10 23:32:18 +00:00
- INSTALLS += locales resources
no-sse2.patch: Install the shared libraries, don't link the sse2 one
2016-01-19 13:55:13 +00:00
+ # install the shared libraries
+ shlibs.CONFIG += no_check_exist
+ shlibs.path = $$shlib_install_path
+ shlibs_sse2.CONFIG += no_check_exist
+ shlibs_sse2.path = $$shlib_install_path/sse2
Update to 5.6.1 * Fri Jun 10 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.1-1 - Update to 5.6.1 - Rebase linux-pri patch (drop the parts already fixed upstream) - Drop backported chimera-nss-init patch, already applied upstream - Rebase no-sse2 patch (the core_module.pro change)
2016-06-10 23:32:18 +00:00
+ INSTALLS += locales resources shlibs shlibs_sse2
no-sse2.patch: Install the shared libraries, don't link the sse2 one
2016-01-19 13:55:13 +00:00
Update to 5.6.1 * Fri Jun 10 2016 Kevin Kofler <Kevin@tigcc.ticalc.org> - 5.6.1-1 - Update to 5.6.1 - Rebase linux-pri patch (drop the parts already fixed upstream) - Drop backported chimera-nss-init patch, already applied upstream - Rebase no-sse2 patch (the core_module.pro change)
2016-06-10 23:32:18 +00:00
!use?(system_icu) {
icu.CONFIG += no_check_exist
Prep webengine 5.9.0
2017-05-31 13:43:14 +00:00
--
2.13.0
Reference in New Issue Copy Permalink