shithub: libvpx

Download patch

ref: 5d0a271ded78c1fe86ca583bef29d16c80d4feba
parent: 6e179dacd00492ff32a14c570a4b9ba673751b85
parent: e10c95dc83c7086665ca77ea39f6c776ff5e9fd8
author: Johann Koenig <johannkoenig@google.com>
date: Sat Nov 5 20:12:12 EDT 2016 

Merge "Update vp9_fdct8x8_quant_ssse3 for highbitdepth"


--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -137,6 +137,7 @@
   add_proto qw/void vp9_quantize_fp_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";
 
   add_proto qw/void vp9_fdct8x8_quant/, "const int16_t *input, int stride, 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/vp9_fdct8x8_quant ssse3/;
 } else {
   add_proto qw/int64_t vp9_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
   specialize qw/vp9_block_error avx2 msa sse2/;
--- a/vp9/encoder/x86/vp9_dct_ssse3.c
+++ b/vp9/encoder/x86/vp9_dct_ssse3.c
@@ -12,14 +12,17 @@
 #include <tmmintrin.h>  // SSSE3
 
 #include "./vp9_rtcd.h"
+#include "./vpx_config.h"
+#include "vpx_dsp/vpx_dsp_common.h"
+#include "vpx_dsp/x86/fdct.h"
 #include "vpx_dsp/x86/inv_txfm_sse2.h"
 #include "vpx_dsp/x86/txfm_common_sse2.h"
 
 void vp9_fdct8x8_quant_ssse3(
-    const int16_t *input, int stride, int16_t *coeff_ptr, intptr_t n_coeffs,
+    const int16_t *input, int stride, 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,
-    int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr, const int16_t *dequant_ptr,
+    tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr,
     uint16_t *eob_ptr, const int16_t *scan_ptr, const int16_t *iscan_ptr) {
   __m128i zero;
   int pass;
@@ -328,15 +331,15 @@
         qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
         qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
 
-        _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), qcoeff0);
-        _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
+        store_tran_low(qcoeff0, qcoeff_ptr + n_coeffs);
+        store_tran_low(qcoeff1, qcoeff_ptr + n_coeffs + 8);
 
         coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
         dequant = _mm_unpackhi_epi64(dequant, dequant);
         coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
 
-        _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), coeff0);
-        _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
+        store_tran_low(coeff0, dqcoeff_ptr + n_coeffs);
+        store_tran_low(coeff1, dqcoeff_ptr + n_coeffs + 8);
       }
 
       {
@@ -398,20 +401,21 @@
           qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
           qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
 
-          _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), qcoeff0);
-          _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
+          store_tran_low(qcoeff0, qcoeff_ptr + n_coeffs);
+          store_tran_low(qcoeff1, qcoeff_ptr + n_coeffs + 8);
 
           coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
           coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
 
-          _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), coeff0);
-          _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
+          store_tran_low(coeff0, dqcoeff_ptr + n_coeffs);
+          store_tran_low(coeff1, dqcoeff_ptr + n_coeffs + 8);
         } else {
-          _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), zero);
-          _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, zero);
+          // Maybe a more efficient way to store 0?
+          store_zero_tran_low(qcoeff_ptr + n_coeffs);
+          store_zero_tran_low(qcoeff_ptr + n_coeffs + 8);
 
-          _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), zero);
-          _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, zero);
+          store_zero_tran_low(dqcoeff_ptr + n_coeffs);
+          store_zero_tran_low(dqcoeff_ptr + n_coeffs + 8);
         }
       }
 
@@ -452,10 +456,10 @@
     }
   } else {
     do {
-      _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), zero);
-      _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, zero);
-      _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), zero);
-      _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, zero);
+      store_zero_tran_low(dqcoeff_ptr + n_coeffs);
+      store_zero_tran_low(dqcoeff_ptr + n_coeffs + 8);
+      store_zero_tran_low(qcoeff_ptr + n_coeffs);
+      store_zero_tran_low(qcoeff_ptr + n_coeffs + 8);
       n_coeffs += 8 * 2;
     } while (n_coeffs < 0);
     *eob_ptr = 0;
--- a/vpx_dsp/x86/fdct.h
+++ b/vpx_dsp/x86/fdct.h
@@ -43,4 +43,15 @@
   _mm_store_si128((__m128i *)(b), a);
 #endif
 }
+
+// Zero fill 8 positions in the output buffer.
+static INLINE void store_zero_tran_low(tran_low_t *a) {
+  const __m128i zero = _mm_setzero_si128();
+#if CONFIG_VP9_HIGHBITDEPTH
+  _mm_store_si128((__m128i *)(a), zero);
+  _mm_store_si128((__m128i *)(a + 4), zero);
+#else
+  _mm_store_si128((__m128i *)(a), zero);
+#endif
+}
 #endif  // VPX_DSP_X86_FDCT_H_