Use 8 bit RGB to Y coefficients for Y and YJ in MMI and MSA.

1. Switch to 8 bit precision.
2. Fix an error in the implementation of MMI and MSA.

About the error:
MMI and MSA implementation for RGBtoY and RGBToYJ used different
precision according to the C implementation( The C version has been
unified in commit fce0fed542001577e6b10f4cf859e0fa1774974e). This
patch unifies the precision to 8 bit for RGBToYJ in MMI and MSA.

Change-Id: Ic6a6e424d27a2f049b0c954f03174192d2beb091
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/2155608
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>

3 files changed, 14 insertions, 19 deletions

diff --git a/source/row_common.cc b/source/row_common.cc
index 67b0c723..2d0f27d4 100644
--- a/source/row_common.cc
+++ b/source/row_common.cc
@@ -26,11 +26,6 @@ extern "C" {
     (defined(_M_IX86) || (defined(_M_X64) && !defined(__clang__)))
 #define LIBYUV_RGB7 1
 #endif
-// mips use 7 bit RGBToY
-#if (!defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)) || \
-    (!defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa))
-#define LIBYUV_RGB7 1
-#endif
 
 // llvm x86 is poor at ternary operator, so use branchless min/max.
 
diff --git a/source/row_mmi.cc b/source/row_mmi.cc
index 9ecafaa2..759c0b52 100644
--- a/source/row_mmi.cc
+++ b/source/row_mmi.cc
@@ -2473,10 +2473,10 @@ void ARGBToYJRow_MMI(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
   uint64_t src, src_hi, src_lo;
   uint64_t dest, dest0, dest1, dest2, dest3;
   uint64_t tmp0, tmp1;
-  const uint64_t shift = 0x07;
-  const uint64_t value = 0x0040;
+  const uint64_t shift = 0x08;
+  const uint64_t value = 0x80;
   const uint64_t mask0 = 0x0;
-  const uint64_t mask1 = 0x00010026004B000FULL;
+  const uint64_t mask1 = 0x0001004D0096001DULL;
 
   __asm__ volatile(
       "1:                                                           \n\t"
@@ -4054,10 +4054,10 @@ void ARGBGrayRow_MMI(const uint8_t* src_argb, uint8_t* dst_argb, int width) {
   uint64_t tmp0, tmp1;
   const uint64_t mask0 = 0x0;
   const uint64_t mask1 = 0x01;
-  const uint64_t mask2 = 0x00400026004B000FULL;
+  const uint64_t mask2 = 0x0080004D0096001DULL;
   const uint64_t mask3 = 0xFF000000FF000000ULL;
   const uint64_t mask4 = ~mask3;
-  const uint64_t shift = 0x07;
+  const uint64_t shift = 0x08;
 
   __asm__ volatile(
       "1:                                                           \n\t"
diff --git a/source/row_msa.cc b/source/row_msa.cc
index 1e410fb6..199e5db3 100644
--- a/source/row_msa.cc
+++ b/source/row_msa.cc
@@ -1412,17 +1412,17 @@ void ARGBGrayRow_MSA(const uint8_t* src_argb, uint8_t* dst_argb, int width) {
   int x;
   v16u8 src0, src1, vec0, vec1, dst0, dst1;
   v8u16 reg0;
-  v16u8 const_0x26 = (v16u8)__msa_ldi_h(0x26);
-  v16u8 const_0x4B0F = (v16u8)__msa_fill_h(0x4B0F);
+  v16u8 const_0x4D = (v16u8)__msa_ldi_h(0x4D);
+  v16u8 const_0x961D = (v16u8)__msa_fill_h(0x961D);
 
   for (x = 0; x < width; x += 8) {
     src0 = (v16u8)__msa_ld_b((v16u8*)src_argb, 0);
     src1 = (v16u8)__msa_ld_b((v16u8*)src_argb, 16);
     vec0 = (v16u8)__msa_pckev_h((v8i16)src1, (v8i16)src0);
     vec1 = (v16u8)__msa_pckod_h((v8i16)src1, (v8i16)src0);
-    reg0 = __msa_dotp_u_h(vec0, const_0x4B0F);
-    reg0 = __msa_dpadd_u_h(reg0, vec1, const_0x26);
-    reg0 = (v8u16)__msa_srari_h((v8i16)reg0, 7);
+    reg0 = __msa_dotp_u_h(vec0, const_0x961D);
+    reg0 = __msa_dpadd_u_h(reg0, vec1, const_0x4D);
+    reg0 = (v8u16)__msa_srari_h((v8i16)reg0, 8);
     vec0 = (v16u8)__msa_ilvev_b((v16i8)reg0, (v16i8)reg0);
     vec1 = (v16u8)__msa_ilvod_b((v16i8)vec1, (v16i8)vec0);
     dst0 = (v16u8)__msa_ilvr_b((v16i8)vec1, (v16i8)vec0);
@@ -2419,16 +2419,16 @@ void SobelXYRow_MSA(const uint8_t* src_sobelx,
 void ARGBToYJRow_MSA(const uint8_t* src_argb0, uint8_t* dst_y, int width) {
   int x;
   v16u8 src0, src1, src2, src3, dst0;
-  v16u8 const_0x4B0F = (v16u8)__msa_fill_h(0x4B0F);
-  v16u8 const_0x26 = (v16u8)__msa_fill_h(0x26);
-  v8u16 const_0x40 = (v8u16)__msa_fill_h(0x40);
+  v16u8 const_0x961D = (v16u8)__msa_fill_h(0x961D);
+  v16u8 const_0x4D = (v16u8)__msa_fill_h(0x4D);
+  v8u16 const_0x80 = (v8u16)__msa_fill_h(0x80);
 
   for (x = 0; x < width; x += 16) {
     src0 = (v16u8)__msa_ld_b((void*)src_argb0, 0);
     src1 = (v16u8)__msa_ld_b((void*)src_argb0, 16);
     src2 = (v16u8)__msa_ld_b((void*)src_argb0, 32);
     src3 = (v16u8)__msa_ld_b((void*)src_argb0, 48);
-    ARGBTOY(src0, src1, src2, src3, const_0x4B0F, const_0x26, const_0x40, 7,
+    ARGBTOY(src0, src1, src2, src3, const_0x961D, const_0x4D, const_0x80, 8,
             dst0);
     ST_UB(dst0, dst_y);
     src_argb0 += 64;