Merge "Optimize vpx_quantize_{b,b_32x32} assembler."

vp10/encoder/context_tree.c

@@ -30,13 +30,13 @@ static void alloc_mode_context(VP10_COMMON *cm, int num_4x4_blk,
   for (i = 0; i < MAX_MB_PLANE; ++i) {
     for (k = 0; k < 3; ++k) {
       CHECK_MEM_ERROR(cm, ctx->coeff[i][k],
-                      vpx_memalign(16, num_pix * sizeof(*ctx->coeff[i][k])));
+                      vpx_memalign(32, num_pix * sizeof(*ctx->coeff[i][k])));
       CHECK_MEM_ERROR(cm, ctx->qcoeff[i][k],
-                      vpx_memalign(16, num_pix * sizeof(*ctx->qcoeff[i][k])));
+                      vpx_memalign(32, num_pix * sizeof(*ctx->qcoeff[i][k])));
       CHECK_MEM_ERROR(cm, ctx->dqcoeff[i][k],
-                      vpx_memalign(16, num_pix * sizeof(*ctx->dqcoeff[i][k])));
+                      vpx_memalign(32, num_pix * sizeof(*ctx->dqcoeff[i][k])));
       CHECK_MEM_ERROR(cm, ctx->eobs[i][k],
-                      vpx_memalign(16, num_blk * sizeof(*ctx->eobs[i][k])));
+                      vpx_memalign(32, num_blk * sizeof(*ctx->eobs[i][k])));
       ctx->coeff_pbuf[i][k]   = ctx->coeff[i][k];
       ctx->qcoeff_pbuf[i][k]  = ctx->qcoeff[i][k];
       ctx->dqcoeff_pbuf[i][k] = ctx->dqcoeff[i][k];

vp9/encoder/vp9_context_tree.c

@@ -30,13 +30,13 @@ static void alloc_mode_context(VP9_COMMON *cm, int num_4x4_blk,
   for (i = 0; i < MAX_MB_PLANE; ++i) {
     for (k = 0; k < 3; ++k) {
       CHECK_MEM_ERROR(cm, ctx->coeff[i][k],
-                      vpx_memalign(16, num_pix * sizeof(*ctx->coeff[i][k])));
+                      vpx_memalign(32, num_pix * sizeof(*ctx->coeff[i][k])));
       CHECK_MEM_ERROR(cm, ctx->qcoeff[i][k],
-                      vpx_memalign(16, num_pix * sizeof(*ctx->qcoeff[i][k])));
+                      vpx_memalign(32, num_pix * sizeof(*ctx->qcoeff[i][k])));
       CHECK_MEM_ERROR(cm, ctx->dqcoeff[i][k],
-                      vpx_memalign(16, num_pix * sizeof(*ctx->dqcoeff[i][k])));
+                      vpx_memalign(32, num_pix * sizeof(*ctx->dqcoeff[i][k])));
       CHECK_MEM_ERROR(cm, ctx->eobs[i][k],
-                      vpx_memalign(16, num_blk * sizeof(*ctx->eobs[i][k])));
+                      vpx_memalign(32, num_blk * sizeof(*ctx->eobs[i][k])));
       ctx->coeff_pbuf[i][k]   = ctx->coeff[i][k];
       ctx->qcoeff_pbuf[i][k]  = ctx->qcoeff[i][k];
       ctx->dqcoeff_pbuf[i][k] = ctx->dqcoeff[i][k];

vpx_dsp/vpx_dsp.mk

@@ -248,7 +248,8 @@ DSP_SRCS-$(HAVE_SSE2)   += x86/highbd_quantize_intrin_sse2.c
 endif
 ifeq ($(ARCH_X86_64),yes)
 ifeq ($(CONFIG_USE_X86INC),yes)
-DSP_SRCS-$(HAVE_SSSE3) += x86/quantize_ssse3_x86_64.asm
+DSP_SRCS-$(HAVE_SSSE3)  += x86/quantize_ssse3_x86_64.asm
+DSP_SRCS-$(HAVE_AVX)    += x86/quantize_avx_x86_64.asm
 endif
 endif
 endif  # CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER

vpx_dsp/vpx_dsp_rtcd_defs.pl

@@ -913,25 +913,19 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
 # Quantization
 #
 if ((vpx_config("CONFIG_VP9_ENCODER") eq "yes") || (vpx_config("CONFIG_VP10_ENCODER") eq "yes")) {
-if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   add_proto qw/void vpx_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-  specialize qw/vpx_quantize_b sse2/, "$ssse3_x86_64_x86inc";
+  specialize qw/vpx_quantize_b sse2/, "$ssse3_x86_64_x86inc", "$avx_x86_64_x86inc";
 
   add_proto qw/void vpx_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-  specialize qw/vpx_quantize_b_32x32/, "$ssse3_x86_64_x86inc";
-
-  add_proto qw/void vpx_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-  specialize qw/vpx_highbd_quantize_b sse2/;
+  specialize qw/vpx_quantize_b_32x32/, "$ssse3_x86_64_x86inc", "$avx_x86_64_x86inc";
 
-  add_proto qw/void vpx_highbd_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-  specialize qw/vpx_highbd_quantize_b_32x32 sse2/;
-} else {
-  add_proto qw/void vpx_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-  specialize qw/vpx_quantize_b sse2/, "$ssse3_x86_64_x86inc";
+  if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
+    add_proto qw/void vpx_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+    specialize qw/vpx_highbd_quantize_b sse2/;
 
-  add_proto qw/void vpx_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
-  specialize qw/vpx_quantize_b_32x32/, "$ssse3_x86_64_x86inc";
-}  # CONFIG_VP9_HIGHBITDEPTH
+    add_proto qw/void vpx_highbd_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
+    specialize qw/vpx_highbd_quantize_b_32x32 sse2/;
+  }  # CONFIG_VP9_HIGHBITDEPTH
 }  # CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
 
 if (vpx_config("CONFIG_ENCODERS") eq "yes") {