Backports highbitdepth accelerations into vp10

Ports the changes in
https://chromium-review.googlesource.com/#/c/302372/3
into vp10.

Change-Id: I334c409f693691227ad16fc703c91899592dd8dc

diff --git a/vp10/common/vp10_rtcd_defs.pl b/vp10/common/vp10_rtcd_defs.pl
index 0d14ad85bacd6ff26808ec4d0e78dc254d80f772..4404701878dab452c48c21986e2c0f096ba06c3f 100644 (file)
--- a/vp10/common/vp10_rtcd_defs.pl
+++ b/vp10/common/vp10_rtcd_defs.pl
@@ -87,65 +87,127 @@ specialize qw/vp10_filter_by_weight8x8 sse2 msa/;
 if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   # Note as optimized versions of these functions are added we need to add a check to ensure
   # that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only.
-  add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
-  specialize qw/vp10_iht4x4_16_add/;
+  if (vpx_config("CONFIG_EMULATE_HARDWARE") eq "yes") {
+    add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
+    specialize qw/vp10_iht4x4_16_add/;
+
+    add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
+    specialize qw/vp10_iht8x8_64_add/;
+
+    add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
+    specialize qw/vp10_iht16x16_256_add/;
+
+    add_proto qw/void vp10_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct4x4/;
+
+    add_proto qw/void vp10_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct4x4_1/;
+
+    add_proto qw/void vp10_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct8x8/;
+
+    add_proto qw/void vp10_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct8x8_1/;
+
+    add_proto qw/void vp10_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct16x16/;
+
+    add_proto qw/void vp10_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct16x16_1/;
+
+    add_proto qw/void vp10_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32/;
+
+    add_proto qw/void vp10_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32_rd/;
+
+    add_proto qw/void vp10_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32_1/;
+
+    add_proto qw/void vp10_highbd_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct4x4/;
+
+    add_proto qw/void vp10_highbd_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct8x8/;
+
+    add_proto qw/void vp10_highbd_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct8x8_1/;
 
-  add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
-  specialize qw/vp10_iht8x8_64_add/;
+    add_proto qw/void vp10_highbd_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct16x16/;
 
-  add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
-  specialize qw/vp10_iht16x16_256_add/;
+    add_proto qw/void vp10_highbd_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct16x16_1/;
 
-  add_proto qw/void vp10_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_fdct4x4 sse2/;
+    add_proto qw/void vp10_highbd_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct32x32/;
+
+    add_proto qw/void vp10_highbd_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct32x32_rd/;
+
+    add_proto qw/void vp10_highbd_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct32x32_1/;
+  } else {
+    add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
+    specialize qw/vp10_iht4x4_16_add sse2/;
+
+    add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
+    specialize qw/vp10_iht8x8_64_add sse2/;
+
+    add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
+    specialize qw/vp10_iht16x16_256_add/;
+
+    add_proto qw/void vp10_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct4x4 sse2/;
 
-  add_proto qw/void vp10_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_fdct4x4_1 sse2/;
+    add_proto qw/void vp10_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct4x4_1 sse2/;
 
-  add_proto qw/void vp10_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_fdct8x8 sse2/;
+    add_proto qw/void vp10_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct8x8 sse2/;
 
-  add_proto qw/void vp10_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_fdct8x8_1 sse2/;
+    add_proto qw/void vp10_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct8x8_1 sse2/;
 
-  add_proto qw/void vp10_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_fdct16x16 sse2/;
+    add_proto qw/void vp10_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct16x16 sse2/;
 
-  add_proto qw/void vp10_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_fdct16x16_1 sse2/;
+    add_proto qw/void vp10_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct16x16_1 sse2/;
 
-  add_proto qw/void vp10_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_fdct32x32 sse2/;
+    add_proto qw/void vp10_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32 sse2/;
 
-  add_proto qw/void vp10_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_fdct32x32_rd sse2/;
+    add_proto qw/void vp10_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32_rd sse2/;
 
-  add_proto qw/void vp10_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_fdct32x32_1 sse2/;
+    add_proto qw/void vp10_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32_1 sse2/;
 
-  add_proto qw/void vp10_highbd_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_highbd_fdct4x4 sse2/;
+    add_proto qw/void vp10_highbd_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct4x4 sse2/;
 
-  add_proto qw/void vp10_highbd_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_highbd_fdct8x8 sse2/;
+    add_proto qw/void vp10_highbd_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct8x8 sse2/;
 
-  add_proto qw/void vp10_highbd_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_highbd_fdct8x8_1/;
+    add_proto qw/void vp10_highbd_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct8x8_1/;
 
-  add_proto qw/void vp10_highbd_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_highbd_fdct16x16 sse2/;
+    add_proto qw/void vp10_highbd_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct16x16 sse2/;
 
-  add_proto qw/void vp10_highbd_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_highbd_fdct16x16_1/;
+    add_proto qw/void vp10_highbd_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct16x16_1/;
 
-  add_proto qw/void vp10_highbd_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_highbd_fdct32x32 sse2/;
+    add_proto qw/void vp10_highbd_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct32x32 sse2/;
 
-  add_proto qw/void vp10_highbd_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_highbd_fdct32x32_rd sse2/;
+    add_proto qw/void vp10_highbd_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct32x32_rd sse2/;
 
-  add_proto qw/void vp10_highbd_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vp10_highbd_fdct32x32_1/;
+    add_proto qw/void vp10_highbd_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_highbd_fdct32x32_1/;
+  }
 } else {
   # Force C versions if CONFIG_EMULATE_HARDWARE is 1
   if (vpx_config("CONFIG_EMULATE_HARDWARE") eq "yes") {

diff --git a/vp10/common/x86/idct_intrin_sse2.c b/vp10/common/x86/idct_intrin_sse2.c
index d58e2635826f50cf43977b0f1198f058a833e01c..a2c674b802fb5a2cb409441b553fe544abd93077 100644 (file)
--- a/vp10/common/x86/idct_intrin_sse2.c
+++ b/vp10/common/x86/idct_intrin_sse2.c
@@ -12,14 +12,14 @@
 #include "vpx_dsp/x86/txfm_common_sse2.h"
 #include "vpx_ports/mem.h"
 
-void vp10_iht4x4_16_add_sse2(const int16_t *input, uint8_t *dest, int stride,
-                            int tx_type) {
+void vp10_iht4x4_16_add_sse2(const tran_low_t *input, uint8_t *dest, int stride,
+                             int tx_type) {
   __m128i in[2];
   const __m128i zero = _mm_setzero_si128();
   const __m128i eight = _mm_set1_epi16(8);
 
-  in[0] = _mm_loadu_si128((const __m128i *)(input));
-  in[1] = _mm_loadu_si128((const __m128i *)(input + 8));
+  in[0] = load_input_data(input);
+  in[1] = load_input_data(input + 8);
 
   switch (tx_type) {
     case 0:  // DCT_DCT
@@ -77,21 +77,21 @@ void vp10_iht4x4_16_add_sse2(const int16_t *input, uint8_t *dest, int stride,
   }
 }
 
-void vp10_iht8x8_64_add_sse2(const int16_t *input, uint8_t *dest, int stride,
-                            int tx_type) {
+void vp10_iht8x8_64_add_sse2(const tran_low_t *input, uint8_t *dest, int stride,
+                             int tx_type) {
   __m128i in[8];
   const __m128i zero = _mm_setzero_si128();
   const __m128i final_rounding = _mm_set1_epi16(1 << 4);
 
   // load input data
-  in[0] = _mm_load_si128((const __m128i *)input);
-  in[1] = _mm_load_si128((const __m128i *)(input + 8 * 1));
-  in[2] = _mm_load_si128((const __m128i *)(input + 8 * 2));
-  in[3] = _mm_load_si128((const __m128i *)(input + 8 * 3));
-  in[4] = _mm_load_si128((const __m128i *)(input + 8 * 4));
-  in[5] = _mm_load_si128((const __m128i *)(input + 8 * 5));
-  in[6] = _mm_load_si128((const __m128i *)(input + 8 * 6));
-  in[7] = _mm_load_si128((const __m128i *)(input + 8 * 7));
+  in[0] = load_input_data(input);
+  in[1] = load_input_data(input + 8 * 1);
+  in[2] = load_input_data(input + 8 * 2);
+  in[3] = load_input_data(input + 8 * 3);
+  in[4] = load_input_data(input + 8 * 4);
+  in[5] = load_input_data(input + 8 * 5);
+  in[6] = load_input_data(input + 8 * 6);
+  in[7] = load_input_data(input + 8 * 7);
 
   switch (tx_type) {
     case 0:  // DCT_DCT
@@ -144,8 +144,8 @@ void vp10_iht8x8_64_add_sse2(const int16_t *input, uint8_t *dest, int stride,
   RECON_AND_STORE(dest + 7 * stride, in[7]);
 }
 
-void vp10_iht16x16_256_add_sse2(const int16_t *input, uint8_t *dest, int stride,
-                               int tx_type) {
+void vp10_iht16x16_256_add_sse2(const tran_low_t *input, uint8_t *dest,
+                                int stride, int tx_type) {
   __m128i in0[16], in1[16];
 
   load_buffer_8x16(input, in0);