diff options
Diffstat (limited to 'third_party/aom/av1/encoder/x86')
| -rw-r--r-- | third_party/aom/av1/encoder/x86/av1_fwd_txfm_sse2.c | 6 | ||||
| -rw-r--r-- | third_party/aom/av1/encoder/x86/cnn_avx2.c | 2 |
2 files changed, 7 insertions, 1 deletions
diff --git a/third_party/aom/av1/encoder/x86/av1_fwd_txfm_sse2.c b/third_party/aom/av1/encoder/x86/av1_fwd_txfm_sse2.c index a4def754b0..31cc37db7a 100644 --- a/third_party/aom/av1/encoder/x86/av1_fwd_txfm_sse2.c +++ b/third_party/aom/av1/encoder/x86/av1_fwd_txfm_sse2.c @@ -2638,6 +2638,11 @@ void av1_lowbd_fwd_txfm2d_16x64_sse2(const int16_t *input, int32_t *output, } } +// Include top-level function only for 32-bit x86, to support Valgrind. +// For normal use, we require SSE4.1, so av1_lowbd_fwd_txfm_sse4_1 will be used +// instead of this function. However, 32-bit Valgrind does not support SSE4.1, +// so we include a fallback to SSE2 to improve performance +#if AOM_ARCH_X86 static FwdTxfm2dFunc fwd_txfm2d_func_ls[TX_SIZES_ALL] = { av1_lowbd_fwd_txfm2d_4x4_sse2, // 4x4 transform av1_lowbd_fwd_txfm2d_8x8_sse2, // 8x8 transform @@ -2671,3 +2676,4 @@ void av1_lowbd_fwd_txfm_sse2(const int16_t *src_diff, tran_low_t *coeff, fwd_txfm2d_func(src_diff, coeff, diff_stride, txfm_param->tx_type, txfm_param->bd); } +#endif // AOM_ARCH_X86 diff --git a/third_party/aom/av1/encoder/x86/cnn_avx2.c b/third_party/aom/av1/encoder/x86/cnn_avx2.c index ee93b3d5a0..9c26a56641 100644 --- a/third_party/aom/av1/encoder/x86/cnn_avx2.c +++ b/third_party/aom/av1/encoder/x86/cnn_avx2.c @@ -466,7 +466,7 @@ static INLINE void cnn_convolve_no_maxpool_padding_valid_layer2_avx2( // As per the layer config set by av1_intra_mode_cnn_partition_cnn_config, // the filter_width and filter_height are equal to 2 for layer >= 1. So // convolution happens at 2x2 for layer >= 1. -void cnn_convolve_no_maxpool_padding_valid_2x2_avx2( +static void cnn_convolve_no_maxpool_padding_valid_2x2_avx2( const float **input, int in_width, int in_height, int in_stride, const CNN_LAYER_CONFIG *const layer_config, float **output, int out_stride, int start_idx, const int cstep, const int channel_step) { |