diff options
Diffstat (limited to 'third_party/aom/aom_dsp/x86/variance_avx2.c')
-rw-r--r-- | third_party/aom/aom_dsp/x86/variance_avx2.c | 26 |
1 files changed, 14 insertions, 12 deletions
diff --git a/third_party/aom/aom_dsp/x86/variance_avx2.c b/third_party/aom/aom_dsp/x86/variance_avx2.c index 046d6f10f8..0f872fc392 100644 --- a/third_party/aom/aom_dsp/x86/variance_avx2.c +++ b/third_party/aom/aom_dsp/x86/variance_avx2.c @@ -518,8 +518,8 @@ void aom_highbd_comp_mask_pred_avx2(uint8_t *comp_pred8, const uint8_t *pred8, } } -uint64_t aom_mse_4xh_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, - int sstride, int h) { +static uint64_t mse_4xh_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, + int sstride, int h) { uint64_t sum = 0; __m128i dst0_4x8, dst1_4x8, dst2_4x8, dst3_4x8, dst_16x8; __m128i src0_4x16, src1_4x16, src2_4x16, src3_4x16; @@ -575,8 +575,9 @@ uint64_t aom_mse_4xh_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, // In src buffer, each 4x4 block in a 32x32 filter block is stored sequentially. // Hence src_blk_stride is same as block width. Whereas dst buffer is a frame // buffer, thus dstride is a frame level stride. -uint64_t aom_mse_4xh_quad_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, - int src_blk_stride, int h) { +static uint64_t mse_4xh_quad_16bit_avx2(uint8_t *dst, int dstride, + uint16_t *src, int src_blk_stride, + int h) { uint64_t sum = 0; __m128i dst0_16x8, dst1_16x8, dst2_16x8, dst3_16x8; __m256i dst0_16x16, dst1_16x16, dst2_16x16, dst3_16x16; @@ -665,8 +666,8 @@ uint64_t aom_mse_4xh_quad_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, return sum; } -uint64_t aom_mse_8xh_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, - int sstride, int h) { +static uint64_t mse_8xh_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, + int sstride, int h) { uint64_t sum = 0; __m128i dst0_8x8, dst1_8x8, dst3_16x8; __m256i src0_8x16, src1_8x16, src_16x16, dst_16x16; @@ -715,8 +716,9 @@ uint64_t aom_mse_8xh_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, // In src buffer, each 8x8 block in a 64x64 filter block is stored sequentially. // Hence src_blk_stride is same as block width. Whereas dst buffer is a frame // buffer, thus dstride is a frame level stride. -uint64_t aom_mse_8xh_dual_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, - int src_blk_stride, int h) { +static uint64_t mse_8xh_dual_16bit_avx2(uint8_t *dst, int dstride, + uint16_t *src, int src_blk_stride, + int h) { uint64_t sum = 0; __m128i dst0_16x8, dst1_16x8; __m256i dst0_16x16, dst1_16x16; @@ -780,8 +782,8 @@ uint64_t aom_mse_wxh_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, assert((w == 8 || w == 4) && (h == 8 || h == 4) && "w=8/4 and h=8/4 must be satisfied"); switch (w) { - case 4: return aom_mse_4xh_16bit_avx2(dst, dstride, src, sstride, h); - case 8: return aom_mse_8xh_16bit_avx2(dst, dstride, src, sstride, h); + case 4: return mse_4xh_16bit_avx2(dst, dstride, src, sstride, h); + case 8: return mse_8xh_16bit_avx2(dst, dstride, src, sstride, h); default: assert(0 && "unsupported width"); return -1; } } @@ -795,8 +797,8 @@ uint64_t aom_mse_16xh_16bit_avx2(uint8_t *dst, int dstride, uint16_t *src, assert((w == 8 || w == 4) && (h == 8 || h == 4) && "w=8/4 and h=8/4 must be satisfied"); switch (w) { - case 4: return aom_mse_4xh_quad_16bit_avx2(dst, dstride, src, w * h, h); - case 8: return aom_mse_8xh_dual_16bit_avx2(dst, dstride, src, w * h, h); + case 4: return mse_4xh_quad_16bit_avx2(dst, dstride, src, w * h, h); + case 8: return mse_8xh_dual_16bit_avx2(dst, dstride, src, w * h, h); default: assert(0 && "unsupported width"); return -1; } } |