diff options
Diffstat (limited to 'vpx_dsp')
| -rw-r--r-- | vpx_dsp/arm/variance_neon.c | 12 | ||||
| -rw-r--r-- | vpx_dsp/x86/inv_txfm_sse2.c | 10 | ||||
| -rw-r--r-- | vpx_dsp/x86/variance_avx2.c | 16 |
3 files changed, 14 insertions, 24 deletions
diff --git a/vpx_dsp/arm/variance_neon.c b/vpx_dsp/arm/variance_neon.c index f469afc4e..b6d7f86a4 100644 --- a/vpx_dsp/arm/variance_neon.c +++ b/vpx_dsp/arm/variance_neon.c @@ -75,7 +75,7 @@ unsigned int vpx_variance8x8_neon(const uint8_t *a, int a_stride, unsigned int *sse) { int sum; variance_neon_w8(a, a_stride, b, b_stride, 8, 8, sse, &sum); - return *sse - (((int64_t)sum * sum) >> 6); // >> 6 = / 8 * 8 + return *sse - ((sum * sum) >> 6); } unsigned int vpx_variance16x16_neon(const uint8_t *a, int a_stride, @@ -83,7 +83,7 @@ unsigned int vpx_variance16x16_neon(const uint8_t *a, int a_stride, unsigned int *sse) { int sum; variance_neon_w8(a, a_stride, b, b_stride, 16, 16, sse, &sum); - return *sse - (((int64_t)sum * sum) >> 8); // >> 8 = / 16 * 16 + return *sse - (((uint32_t)((int64_t)sum * sum)) >> 8); } unsigned int vpx_variance32x32_neon(const uint8_t *a, int a_stride, @@ -91,7 +91,7 @@ unsigned int vpx_variance32x32_neon(const uint8_t *a, int a_stride, unsigned int *sse) { int sum; variance_neon_w8(a, a_stride, b, b_stride, 32, 32, sse, &sum); - return *sse - (((int64_t)sum * sum) >> 10); // >> 10 = / 32 * 32 + return *sse - (unsigned int)(((int64_t)sum * sum) >> 10); } unsigned int vpx_variance32x64_neon(const uint8_t *a, int a_stride, @@ -104,7 +104,7 @@ unsigned int vpx_variance32x64_neon(const uint8_t *a, int a_stride, 32, 32, &sse2, &sum2); *sse = sse1 + sse2; sum1 += sum2; - return *sse - (((int64_t)sum1 * sum1) >> 11); // >> 11 = / 32 * 64 + return *sse - (unsigned int)(((int64_t)sum1 * sum1) >> 11); } unsigned int vpx_variance64x32_neon(const uint8_t *a, int a_stride, @@ -117,7 +117,7 @@ unsigned int vpx_variance64x32_neon(const uint8_t *a, int a_stride, 64, 16, &sse2, &sum2); *sse = sse1 + sse2; sum1 += sum2; - return *sse - (((int64_t)sum1 * sum1) >> 11); // >> 11 = / 32 * 64 + return *sse - (unsigned int)(((int64_t)sum1 * sum1) >> 11); } unsigned int vpx_variance64x64_neon(const uint8_t *a, int a_stride, @@ -141,7 +141,7 @@ unsigned int vpx_variance64x64_neon(const uint8_t *a, int a_stride, b_stride, 64, 16, &sse2, &sum2); *sse = sse1 + sse2; sum1 += sum2; - return *sse - (((int64_t)sum1 * sum1) >> 12); // >> 12 = / 64 * 64 + return *sse - (unsigned int)(((int64_t)sum1 * sum1) >> 12); } unsigned int vpx_variance16x8_neon(const unsigned char *src_ptr, diff --git a/vpx_dsp/x86/inv_txfm_sse2.c b/vpx_dsp/x86/inv_txfm_sse2.c index 330ae8d6a..cb56ad078 100644 --- a/vpx_dsp/x86/inv_txfm_sse2.c +++ b/vpx_dsp/x86/inv_txfm_sse2.c @@ -3066,17 +3066,7 @@ void vpx_idct32x32_34_add_sse2(const tran_low_t *input, uint8_t *dest, in[6] = load_input_data(input + 192); in[7] = load_input_data(input + 224); - for (i = 8; i < 32; ++i) { - in[i] = _mm_setzero_si128(); - } - array_transpose_8x8(in, in); - // TODO(hkuang): Following transposes are unnecessary. But remove them will - // lead to performance drop on some devices. - array_transpose_8x8(in + 8, in + 8); - array_transpose_8x8(in + 16, in + 16); - array_transpose_8x8(in + 24, in + 24); - IDCT32_34 // 1_D: Store 32 intermediate results for each 8x32 block. diff --git a/vpx_dsp/x86/variance_avx2.c b/vpx_dsp/x86/variance_avx2.c index 7bc2693cf..8428e0520 100644 --- a/vpx_dsp/x86/variance_avx2.c +++ b/vpx_dsp/x86/variance_avx2.c @@ -61,7 +61,7 @@ unsigned int vpx_variance32x16_avx2(const uint8_t *src, int src_stride, int sum; variance_avx2(src, src_stride, ref, ref_stride, 32, 16, sse, &sum, vpx_get32x32var_avx2, 32); - return *sse - (((int64_t)sum * sum) >> 9); + return *sse - (uint32_t)(((int64_t)sum * sum) >> 9); } unsigned int vpx_variance32x32_avx2(const uint8_t *src, int src_stride, @@ -70,7 +70,7 @@ unsigned int vpx_variance32x32_avx2(const uint8_t *src, int src_stride, int sum; variance_avx2(src, src_stride, ref, ref_stride, 32, 32, sse, &sum, vpx_get32x32var_avx2, 32); - return *sse - (((int64_t)sum * sum) >> 10); + return *sse - (uint32_t)(((int64_t)sum * sum) >> 10); } unsigned int vpx_variance64x64_avx2(const uint8_t *src, int src_stride, @@ -79,7 +79,7 @@ unsigned int vpx_variance64x64_avx2(const uint8_t *src, int src_stride, int sum; variance_avx2(src, src_stride, ref, ref_stride, 64, 64, sse, &sum, vpx_get32x32var_avx2, 32); - return *sse - (((int64_t)sum * sum) >> 12); + return *sse - (uint32_t)(((int64_t)sum * sum) >> 12); } unsigned int vpx_variance64x32_avx2(const uint8_t *src, int src_stride, @@ -88,7 +88,7 @@ unsigned int vpx_variance64x32_avx2(const uint8_t *src, int src_stride, int sum; variance_avx2(src, src_stride, ref, ref_stride, 64, 32, sse, &sum, vpx_get32x32var_avx2, 32); - return *sse - (((int64_t)sum * sum) >> 11); + return *sse - (uint32_t)(((int64_t)sum * sum) >> 11); } unsigned int vpx_sub_pixel_variance32xh_avx2(const uint8_t *src, int src_stride, @@ -115,7 +115,7 @@ unsigned int vpx_sub_pixel_variance64x64_avx2(const uint8_t *src, dst + 32, dst_stride, 64, &sse2); const int se = se1 + se2; *sse = sse1 + sse2; - return *sse - (((int64_t)se * se) >> 12); + return *sse - (uint32_t)(((int64_t)se * se) >> 12); } unsigned int vpx_sub_pixel_variance32x32_avx2(const uint8_t *src, @@ -125,7 +125,7 @@ unsigned int vpx_sub_pixel_variance32x32_avx2(const uint8_t *src, unsigned int *sse) { const int se = vpx_sub_pixel_variance32xh_avx2( src, src_stride, x_offset, y_offset, dst, dst_stride, 32, sse); - return *sse - (((int64_t)se * se) >> 10); + return *sse - (uint32_t)(((int64_t)se * se) >> 10); } unsigned int vpx_sub_pixel_avg_variance64x64_avx2( @@ -142,7 +142,7 @@ unsigned int vpx_sub_pixel_avg_variance64x64_avx2( *sse = sse1 + sse2; - return *sse - (((int64_t)se * se) >> 12); + return *sse - (uint32_t)(((int64_t)se * se) >> 12); } unsigned int vpx_sub_pixel_avg_variance32x32_avx2( @@ -151,5 +151,5 @@ unsigned int vpx_sub_pixel_avg_variance32x32_avx2( // Process 32 elements in parallel. const int se = vpx_sub_pixel_avg_variance32xh_avx2( src, src_stride, x_offset, y_offset, dst, dst_stride, sec, 32, 32, sse); - return *sse - (((int64_t)se * se) >> 10); + return *sse - (uint32_t)(((int64_t)se * se) >> 10); } |