more fixes
This commit is contained in:
Tom Callaway
parent
fd279120cf
commit
529d54268d
|
|
@ -1,14 +1,14 @@
|
|||
diff -up chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctfst-neon.c.arm-incompatible-ints chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctfst-neon.c
|
||||
--- chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctfst-neon.c.arm-incompatible-ints 2020-07-22 15:17:07.171430022 -0400
|
||||
+++ chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctfst-neon.c 2020-07-22 15:31:40.068296015 -0400
|
||||
diff -up chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctfst-neon.c.el8-arm-incompatible-ints chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctfst-neon.c
|
||||
--- chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctfst-neon.c.el8-arm-incompatible-ints 2020-07-13 14:43:44.000000000 -0400
|
||||
+++ chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctfst-neon.c 2020-07-22 23:14:37.425855588 -0400
|
||||
@@ -84,8 +84,8 @@ void jsimd_idct_ifast_neon(void *dct_tab
|
||||
bitmap = vorrq_s16(bitmap, row6);
|
||||
bitmap = vorrq_s16(bitmap, row7);
|
||||
|
||||
- int64_t left_ac_bitmap = vreinterpret_s64_s16(vget_low_s16(bitmap));
|
||||
- int64_t right_ac_bitmap = vreinterpret_s64_s16(vget_high_s16(bitmap));
|
||||
+ int64x1_t left_ac_bitmap = vreinterpret_s64_s16(vget_low_s16(bitmap));
|
||||
+ int64x1_t right_ac_bitmap = vreinterpret_s64_s16(vget_high_s16(bitmap));
|
||||
+ int64_t left_ac_bitmap = (int64_t)vreinterpret_s64_s16(vget_low_s16(bitmap));
|
||||
+ int64_t right_ac_bitmap = (int64_t)vreinterpret_s64_s16(vget_high_s16(bitmap));
|
||||
|
||||
if (left_ac_bitmap == 0 && right_ac_bitmap == 0) {
|
||||
/* All AC coefficients are zero. */
|
||||
|
|
@ -30,39 +30,90 @@ diff -up chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctfs
|
|||
|
||||
/* Transpose block ready for store. */
|
||||
uint32x4x2_t cols_0415 = vzipq_u32(vreinterpretq_u32_u8(cols_01),
|
||||
diff -up chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctint-neon.c.arm-incompatible-ints chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctint-neon.c
|
||||
--- chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctint-neon.c.arm-incompatible-ints 2020-07-22 16:28:13.318222388 -0400
|
||||
+++ chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctint-neon.c 2020-07-22 16:32:39.578126196 -0400
|
||||
diff -up chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctint-neon.c.el8-arm-incompatible-ints chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctint-neon.c
|
||||
--- chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctint-neon.c.el8-arm-incompatible-ints 2020-07-13 14:43:44.000000000 -0400
|
||||
+++ chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctint-neon.c 2020-07-22 23:33:26.502244720 -0400
|
||||
@@ -214,13 +214,13 @@ void jsimd_idct_islow_neon(void *dct_tab
|
||||
int16x4_t bitmap = vorr_s16(row7, row6);
|
||||
bitmap = vorr_s16(bitmap, row5);
|
||||
bitmap = vorr_s16(bitmap, row4);
|
||||
- int64_t bitmap_rows_4567 = vreinterpret_s64_s16(bitmap);
|
||||
+ int64x1_t bitmap_rows_4567 = vreinterpret_s64_s16(bitmap);
|
||||
+ int64_t bitmap_rows_4567 = (int64_t)vreinterpret_s64_s16(bitmap);
|
||||
|
||||
if (bitmap_rows_4567 == 0) {
|
||||
bitmap = vorr_s16(bitmap, row3);
|
||||
bitmap = vorr_s16(bitmap, row2);
|
||||
bitmap = vorr_s16(bitmap, row1);
|
||||
- int64_t left_ac_bitmap = vreinterpret_s64_s16(bitmap);
|
||||
+ int64x1_t left_ac_bitmap = vreinterpret_s64_s16(bitmap);
|
||||
+ int64_t left_ac_bitmap = (int64_t)vreinterpret_s64_s16(bitmap);
|
||||
|
||||
if (left_ac_bitmap == 0) {
|
||||
int16x4_t dcval = vshl_n_s16(vmul_s16(row0, quant_row0), PASS1_BITS);
|
||||
@@ -270,10 +270,10 @@ void jsimd_idct_islow_neon(void *dct_tab
|
||||
@@ -266,18 +266,18 @@ void jsimd_idct_islow_neon(void *dct_tab
|
||||
bitmap = vorr_s16(row7, row6);
|
||||
bitmap = vorr_s16(bitmap, row5);
|
||||
bitmap = vorr_s16(bitmap, row4);
|
||||
- bitmap_rows_4567 = vreinterpret_s64_s16(bitmap);
|
||||
+ bitmap_rows_4567 = (int64_t)vreinterpret_s64_s16(bitmap);
|
||||
bitmap = vorr_s16(bitmap, row3);
|
||||
bitmap = vorr_s16(bitmap, row2);
|
||||
bitmap = vorr_s16(bitmap, row1);
|
||||
- int64_t right_ac_bitmap = vreinterpret_s64_s16(bitmap);
|
||||
+ int64x1_t right_ac_bitmap = vreinterpret_s64_s16(bitmap);
|
||||
+ int64_t right_ac_bitmap = (int64_t)vreinterpret_s64_s16(bitmap);
|
||||
|
||||
/* Initialise to non-zero value: defaults to regular second pass. */
|
||||
- int64_t right_ac_dc_bitmap = 1;
|
||||
+ int64x1_t right_ac_dc_bitmap = 1;
|
||||
int64_t right_ac_dc_bitmap = 1;
|
||||
|
||||
if (right_ac_bitmap == 0) {
|
||||
bitmap = vorr_s16(bitmap, row0);
|
||||
@@ -630,10 +630,10 @@ static inline void jsimd_idct_islow_pass
|
||||
- right_ac_dc_bitmap = vreinterpret_s64_s16(bitmap);
|
||||
+ right_ac_dc_bitmap = (int64_t)vreinterpret_s64_s16(bitmap);
|
||||
|
||||
if (right_ac_dc_bitmap != 0) {
|
||||
int16x4_t dcval = vshl_n_s16(vmul_s16(row0, quant_row0), PASS1_BITS);
|
||||
@@ -346,7 +346,13 @@ static inline void jsimd_idct_islow_pass
|
||||
int16_t *workspace_2)
|
||||
{
|
||||
/* Load constants for IDCT calculation. */
|
||||
+#if defined(__aarch64__) || defined(__ARM64__) || defined(_M_ARM64)
|
||||
const int16x4x3_t consts = vld1_s16_x3(jsimd_idct_islow_neon_consts);
|
||||
+#else
|
||||
+ const int16x4x3_t consts = { vld1_s16(jsimd_idct_islow_neon_consts),
|
||||
+ vld1_s16(jsimd_idct_islow_neon_consts + 4),
|
||||
+ vld1_s16(jsimd_idct_islow_neon_consts + 8) };
|
||||
+#endif
|
||||
|
||||
/* Even part. */
|
||||
int16x4_t z2_s16 = vmul_s16(row2, quant_row2);
|
||||
@@ -463,7 +469,13 @@ static inline void jsimd_idct_islow_pass
|
||||
int16_t *workspace_2)
|
||||
{
|
||||
/* Load constants for IDCT computation. */
|
||||
+#if defined(__aarch64__) || defined(__ARM64__) || defined(_M_ARM64)
|
||||
const int16x4x3_t consts = vld1_s16_x3(jsimd_idct_islow_neon_consts);
|
||||
+#else
|
||||
+ const int16x4x3_t consts = { vld1_s16(jsimd_idct_islow_neon_consts),
|
||||
+ vld1_s16(jsimd_idct_islow_neon_consts + 4),
|
||||
+ vld1_s16(jsimd_idct_islow_neon_consts + 8) };
|
||||
+#endif
|
||||
|
||||
/* Even part. */
|
||||
int16x4_t z2_s16 = vmul_s16(row2, quant_row2);
|
||||
@@ -537,7 +549,13 @@ static inline void jsimd_idct_islow_pass
|
||||
unsigned buf_offset)
|
||||
{
|
||||
/* Load constants for IDCT computation. */
|
||||
+#if defined(__aarch64__) || defined(__ARM64__) || defined(_M_ARM64)
|
||||
const int16x4x3_t consts = vld1_s16_x3(jsimd_idct_islow_neon_consts);
|
||||
+#else
|
||||
+ const int16x4x3_t consts = { vld1_s16(jsimd_idct_islow_neon_consts),
|
||||
+ vld1_s16(jsimd_idct_islow_neon_consts + 4),
|
||||
+ vld1_s16(jsimd_idct_islow_neon_consts + 8) };
|
||||
+#endif
|
||||
|
||||
/* Even part. */
|
||||
int16x4_t z2_s16 = vld1_s16(workspace + 2 * DCTSIZE / 2);
|
||||
@@ -630,10 +648,10 @@ static inline void jsimd_idct_islow_pass
|
||||
int8x8_t cols_46_s8 = vqrshrn_n_s16(cols_46_s16, DESCALE_P2 - 16);
|
||||
int8x8_t cols_57_s8 = vqrshrn_n_s16(cols_57_s16, DESCALE_P2 - 16);
|
||||
/* Clamp to range [0-255]. */
|
||||
|
|
@ -70,14 +121,28 @@ diff -up chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctin
|
|||
- uint8x8_t cols_13_u8 = vadd_u8(cols_13_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
- uint8x8_t cols_46_u8 = vadd_u8(cols_46_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
- uint8x8_t cols_57_u8 = vadd_u8(cols_57_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_02_u8 = vadd_u8((uint8x8_t)cols_02_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_13_u8 = vadd_u8((uint8x8_t)cols_13_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_46_u8 = vadd_u8((uint8x8_t)cols_46_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_57_u8 = vadd_u8((uint8x8_t)cols_57_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_02_u8 = vadd_u8(vreinterpret_u8_s8(cols_02_s8), vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_13_u8 = vadd_u8(vreinterpret_u8_s8(cols_13_s8), vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_46_u8 = vadd_u8(vreinterpret_u8_s8(cols_46_s8), vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_57_u8 = vadd_u8(vreinterpret_u8_s8(cols_57_s8), vdup_n_u8(CENTERJSAMPLE));
|
||||
|
||||
/* Transpose 4x8 block and store to memory. */
|
||||
/* Zipping adjacent columns together allows us to store 16-bit elements. */
|
||||
@@ -722,10 +722,10 @@ static inline void jsimd_idct_islow_pass
|
||||
@@ -671,7 +689,13 @@ static inline void jsimd_idct_islow_pass
|
||||
unsigned buf_offset)
|
||||
{
|
||||
/* Load constants for IDCT computation. */
|
||||
+#if defined(__aarch64__) || defined(__ARM64__) || defined(_M_ARM64)
|
||||
const int16x4x3_t consts = vld1_s16_x3(jsimd_idct_islow_neon_consts);
|
||||
+#else
|
||||
+ const int16x4x3_t consts = { vld1_s16(jsimd_idct_islow_neon_consts),
|
||||
+ vld1_s16(jsimd_idct_islow_neon_consts + 4),
|
||||
+ vld1_s16(jsimd_idct_islow_neon_consts + 8) };
|
||||
+#endif
|
||||
|
||||
/* Even part. */
|
||||
int16x4_t z2_s16 = vld1_s16(workspace + 2 * DCTSIZE / 2);
|
||||
@@ -722,10 +746,10 @@ static inline void jsimd_idct_islow_pass
|
||||
int8x8_t cols_46_s8 = vqrshrn_n_s16(cols_46_s16, DESCALE_P2 - 16);
|
||||
int8x8_t cols_57_s8 = vqrshrn_n_s16(cols_57_s16, DESCALE_P2 - 16);
|
||||
/* Clamp to range [0-255]. */
|
||||
|
|
@ -85,10 +150,43 @@ diff -up chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctin
|
|||
- uint8x8_t cols_13_u8 = vadd_u8(cols_13_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
- uint8x8_t cols_46_u8 = vadd_u8(cols_46_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
- uint8x8_t cols_57_u8 = vadd_u8(cols_57_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_02_u8 = vadd_u8((uint8x8_t)cols_02_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_13_u8 = vadd_u8((uint8x8_t)cols_13_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_46_u8 = vadd_u8((uint8x8_t)cols_46_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_57_u8 = vadd_u8((uint8x8_t)cols_57_s8, vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_02_u8 = vadd_u8(vreinterpret_u8_s8(cols_02_s8), vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_13_u8 = vadd_u8(vreinterpret_u8_s8(cols_13_s8), vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_46_u8 = vadd_u8(vreinterpret_u8_s8(cols_46_s8), vdup_n_u8(CENTERJSAMPLE));
|
||||
+ uint8x8_t cols_57_u8 = vadd_u8(vreinterpret_u8_s8(cols_57_s8), vdup_n_u8(CENTERJSAMPLE));
|
||||
|
||||
/* Transpose 4x8 block and store to memory. */
|
||||
/* Zipping adjacent columns together allow us to store 16-bit elements. */
|
||||
diff -up chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctred-neon.c.el8-arm-incompatible-ints chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctred-neon.c
|
||||
--- chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctred-neon.c.el8-arm-incompatible-ints 2020-07-22 23:16:15.856337031 -0400
|
||||
+++ chromium-84.0.4147.89/third_party/libjpeg_turbo/simd/arm/common/jidctred-neon.c 2020-07-22 23:28:15.154109493 -0400
|
||||
@@ -217,11 +217,17 @@ void jsimd_idct_4x4_neon(void *dct_table
|
||||
bitmap = vorrq_s16(bitmap, row6);
|
||||
bitmap = vorrq_s16(bitmap, row7);
|
||||
|
||||
- int64_t left_ac_bitmap = vreinterpret_s64_s16(vget_low_s16(bitmap));
|
||||
- int64_t right_ac_bitmap = vreinterpret_s64_s16(vget_high_s16(bitmap));
|
||||
+ int64_t left_ac_bitmap = (int64_t)vreinterpret_s64_s16(vget_low_s16(bitmap));
|
||||
+ int64_t right_ac_bitmap = (int64_t)vreinterpret_s64_s16(vget_high_s16(bitmap));
|
||||
|
||||
/* Load constants for IDCT computation. */
|
||||
+#if defined(__aarch64__) || defined(__ARM64__) || defined(_M_ARM64)
|
||||
const int16x4x3_t consts = vld1_s16_x3(jsimd_idct_4x4_neon_consts);
|
||||
+#else
|
||||
+ const int16x4x3_t consts = { vld1_s16(jsimd_idct_4x4_neon_consts),
|
||||
+ vld1_s16(jsimd_idct_4x4_neon_consts + 4),
|
||||
+ vld1_s16(jsimd_idct_4x4_neon_consts + 8) };
|
||||
+#endif
|
||||
|
||||
if (left_ac_bitmap == 0 && right_ac_bitmap == 0) {
|
||||
/* All AC coefficients are zero. */
|
||||
@@ -452,7 +458,8 @@ void jsimd_idct_4x4_neon(void *dct_table
|
||||
/* Interleaving store completes the transpose. */
|
||||
uint8x8x2_t output_0123 = vzip_u8(vqmovun_s16(output_cols_02),
|
||||
vqmovun_s16(output_cols_13));
|
||||
- uint16x4x2_t output_01_23 = { output_0123.val[0], output_0123.val[1] };
|
||||
+ uint16x4x2_t output_01_23 = { vreinterpret_u16_u8(output_0123.val[0]),
|
||||
+ vreinterpret_u16_u8(output_0123.val[1]) };
|
||||
|
||||
/* Store 4x4 block to memory. */
|
||||
JSAMPROW outptr0 = output_buf[0] + output_col;
|
||||
|
|
|
|||
Loading…
Reference in New Issue