Add mips msa vpx hadamard functions

average improvement ~4x-5x

Change-Id: I167132d894c04fa85dda8dde7906ff9c61b3a65d

2 files changed, 198 insertions, 2 deletions

diff --git a/vpx_dsp/mips/avg_msa.c b/vpx_dsp/mips/avg_msa.c
index 52a24ed37..48e3cf0bb 100644
--- a/vpx_dsp/mips/avg_msa.c
+++ b/vpx_dsp/mips/avg_msa.c
@@ -54,3 +54,199 @@ uint32_t vpx_avg_4x4_msa(const uint8_t *src, int32_t src_stride) {
 
   return sum_out;
 }
+
+void vpx_hadamard_8x8_msa(const int16_t *src, int src_stride, int16_t *dst) {
+  v8i16 src0, src1, src2, src3, src4, src5, src6, src7;
+  v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+
+  LD_SH8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+  BUTTERFLY_8(src0, src2, src4, src6, src7, src5, src3, src1, tmp0, tmp2, tmp4,
+              tmp6, tmp7, tmp5, tmp3, tmp1);
+  BUTTERFLY_8(tmp0, tmp1, tmp4, tmp5, tmp7, tmp6, tmp3, tmp2, src0, src1, src4,
+              src5, src7, src6, src3, src2);
+  BUTTERFLY_8(src0, src1, src2, src3, src7, src6, src5, src4, tmp0, tmp7, tmp3,
+              tmp4, tmp5, tmp1, tmp6, tmp2);
+  TRANSPOSE8x8_SH_SH(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, src0, src1,
+                     src2, src3, src4, src5, src6, src7);
+  BUTTERFLY_8(src0, src2, src4, src6, src7, src5, src3, src1, tmp0, tmp2, tmp4,
+              tmp6, tmp7, tmp5, tmp3, tmp1);
+  BUTTERFLY_8(tmp0, tmp1, tmp4, tmp5, tmp7, tmp6, tmp3, tmp2, src0, src1, src4,
+              src5, src7, src6, src3, src2);
+  BUTTERFLY_8(src0, src1, src2, src3, src7, src6, src5, src4, tmp0, tmp7, tmp3,
+              tmp4, tmp5, tmp1, tmp6, tmp2);
+  TRANSPOSE8x8_SH_SH(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, src0, src1,
+                     src2, src3, src4, src5, src6, src7);
+  ST_SH8(src0, src1, src2, src3, src4, src5, src6, src7, dst, 8);
+}
+
+void vpx_hadamard_16x16_msa(const int16_t *src, int src_stride, int16_t *dst) {
+  v8i16 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+  v8i16 src11, src12, src13, src14, src15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+  v8i16 tmp6, tmp7, tmp8, tmp9, tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
+  v8i16 res0, res1, res2, res3, res4, res5, res6, res7;
+
+  LD_SH2(src, 8, src0, src8);
+  src += src_stride;
+  LD_SH2(src, 8, src1, src9);
+  src += src_stride;
+  LD_SH2(src, 8, src2, src10);
+  src += src_stride;
+  LD_SH2(src, 8, src3, src11);
+  src += src_stride;
+  LD_SH2(src, 8, src4, src12);
+  src += src_stride;
+  LD_SH2(src, 8, src5, src13);
+  src += src_stride;
+  LD_SH2(src, 8, src6, src14);
+  src += src_stride;
+  LD_SH2(src, 8, src7, src15);
+  src += src_stride;
+
+  BUTTERFLY_8(src0, src2, src4, src6, src7, src5, src3, src1, tmp0, tmp2, tmp4,
+              tmp6, tmp7, tmp5, tmp3, tmp1);
+  BUTTERFLY_8(src8, src10, src12, src14, src15, src13, src11, src9, tmp8, tmp10,
+              tmp12, tmp14, tmp15, tmp13, tmp11, tmp9);
+
+  BUTTERFLY_8(tmp0, tmp1, tmp4, tmp5, tmp7, tmp6, tmp3, tmp2, src0, src1, src4,
+              src5, src7, src6, src3, src2);
+  BUTTERFLY_8(src0, src1, src2, src3, src7, src6, src5, src4, tmp0, tmp7, tmp3,
+              tmp4, tmp5, tmp1, tmp6, tmp2);
+  TRANSPOSE8x8_SH_SH(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, src0, src1,
+                     src2, src3, src4, src5, src6, src7);
+  BUTTERFLY_8(src0, src2, src4, src6, src7, src5, src3, src1, tmp0, tmp2, tmp4,
+              tmp6, tmp7, tmp5, tmp3, tmp1);
+  BUTTERFLY_8(tmp0, tmp1, tmp4, tmp5, tmp7, tmp6, tmp3, tmp2, src0, src1, src4,
+              src5, src7, src6, src3, src2);
+  BUTTERFLY_8(src0, src1, src2, src3, src7, src6, src5, src4, tmp0, tmp7, tmp3,
+              tmp4, tmp5, tmp1, tmp6, tmp2);
+  TRANSPOSE8x8_SH_SH(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, src0, src1,
+                     src2, src11, src4, src5, src6, src7);
+  ST_SH8(src0, src1, src2, src11, src4, src5, src6, src7, dst, 8);
+
+  BUTTERFLY_8(tmp8, tmp9, tmp12, tmp13, tmp15, tmp14, tmp11, tmp10, src8, src9,
+              src12, src13, src15, src14, src11, src10);
+  BUTTERFLY_8(src8, src9, src10, src11, src15, src14, src13, src12, tmp8, tmp15,
+              tmp11, tmp12, tmp13, tmp9, tmp14, tmp10);
+  TRANSPOSE8x8_SH_SH(tmp8, tmp9, tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, src8,
+                     src9, src10, src11, src12, src13, src14, src15);
+  BUTTERFLY_8(src8, src10, src12, src14, src15, src13, src11, src9, tmp8, tmp10,
+              tmp12, tmp14, tmp15, tmp13, tmp11, tmp9);
+  BUTTERFLY_8(tmp8, tmp9, tmp12, tmp13, tmp15, tmp14, tmp11, tmp10, src8, src9,
+              src12, src13, src15, src14, src11, src10);
+  BUTTERFLY_8(src8, src9, src10, src11, src15, src14, src13, src12, tmp8, tmp15,
+              tmp11, tmp12, tmp13, tmp9, tmp14, tmp10);
+  TRANSPOSE8x8_SH_SH(tmp8, tmp9, tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, res0,
+                     res1, res2, res3, res4, res5, res6, res7);
+
+  LD_SH2(src, 8, src0, src8);
+  src += src_stride;
+  LD_SH2(src, 8, src1, src9);
+  src += src_stride;
+  LD_SH2(src, 8, src2, src10);
+  src += src_stride;
+  LD_SH2(src, 8, src3, src11);
+  src += src_stride;
+
+  ST_SH8(res0, res1, res2, res3, res4, res5, res6, res7, dst + 64, 8);
+
+  LD_SH2(src, 8, src4, src12);
+  src += src_stride;
+  LD_SH2(src, 8, src5, src13);
+  src += src_stride;
+  LD_SH2(src, 8, src6, src14);
+  src += src_stride;
+  LD_SH2(src, 8, src7, src15);
+  src += src_stride;
+
+  BUTTERFLY_8(src0, src2, src4, src6, src7, src5, src3, src1, tmp0, tmp2, tmp4,
+              tmp6, tmp7, tmp5, tmp3, tmp1);
+  BUTTERFLY_8(src8, src10, src12, src14, src15, src13, src11, src9, tmp8, tmp10,
+              tmp12, tmp14, tmp15, tmp13, tmp11, tmp9);
+
+  BUTTERFLY_8(tmp0, tmp1, tmp4, tmp5, tmp7, tmp6, tmp3, tmp2, src0, src1, src4,
+              src5, src7, src6, src3, src2);
+  BUTTERFLY_8(src0, src1, src2, src3, src7, src6, src5, src4, tmp0, tmp7, tmp3,
+              tmp4, tmp5, tmp1, tmp6, tmp2);
+  TRANSPOSE8x8_SH_SH(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, src0, src1,
+                     src2, src3, src4, src5, src6, src7);
+  BUTTERFLY_8(src0, src2, src4, src6, src7, src5, src3, src1, tmp0, tmp2, tmp4,
+              tmp6, tmp7, tmp5, tmp3, tmp1);
+  BUTTERFLY_8(tmp0, tmp1, tmp4, tmp5, tmp7, tmp6, tmp3, tmp2, src0, src1, src4,
+              src5, src7, src6, src3, src2);
+  BUTTERFLY_8(src0, src1, src2, src3, src7, src6, src5, src4, tmp0, tmp7, tmp3,
+              tmp4, tmp5, tmp1, tmp6, tmp2);
+  TRANSPOSE8x8_SH_SH(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, src0, src1,
+                     src2, src3, src4, src5, src6, src7);
+  ST_SH8(src0, src1, src2, src3, src4, src5, src6, src7, dst + 2 * 64, 8);
+
+  BUTTERFLY_8(tmp8, tmp9, tmp12, tmp13, tmp15, tmp14, tmp11, tmp10, src8, src9,
+              src12, src13, src15, src14, src11, src10);
+  BUTTERFLY_8(src8, src9, src10, src11, src15, src14, src13, src12, tmp8, tmp15,
+              tmp11, tmp12, tmp13, tmp9, tmp14, tmp10);
+  TRANSPOSE8x8_SH_SH(tmp8, tmp9, tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, src8,
+                     src9, src10, src11, src12, src13, src14, src15);
+  BUTTERFLY_8(src8, src10, src12, src14, src15, src13, src11, src9, tmp8, tmp10,
+              tmp12, tmp14, tmp15, tmp13, tmp11, tmp9);
+  BUTTERFLY_8(tmp8, tmp9, tmp12, tmp13, tmp15, tmp14, tmp11, tmp10, src8, src9,
+              src12, src13, src15, src14, src11, src10);
+  BUTTERFLY_8(src8, src9, src10, src11, src15, src14, src13, src12, tmp8, tmp15,
+              tmp11, tmp12, tmp13, tmp9, tmp14, tmp10);
+  TRANSPOSE8x8_SH_SH(tmp8, tmp9, tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, res0,
+                     res1, res2, res3, res4, res5, res6, res7);
+  ST_SH8(res0, res1, res2, res3, res4, res5, res6, res7, dst + 3 * 64, 8);
+
+  LD_SH4(dst, 64, src0, src1, src2, src3);
+  LD_SH4(dst + 8, 64, src4, src5, src6, src7);
+
+  BUTTERFLY_8(src0, src2, src4, src6, src7, src5, src3, src1, tmp0, tmp2, tmp4,
+              tmp6, tmp7, tmp5, tmp3, tmp1);
+  SRA_4V(tmp0, tmp1, tmp2, tmp3, 1);
+  SRA_4V(tmp4, tmp5, tmp6, tmp7, 1);
+  BUTTERFLY_8(tmp0, tmp1, tmp4, tmp5, tmp7, tmp6, tmp3, tmp2, src0, src1, src4,
+              src5, src7, src6, src3, src2);
+
+  ST_SH4(src0, src1, src2, src3, dst, 64);
+  ST_SH4(src4, src5, src6, src7, dst + 8, 64);
+  dst += 16;
+
+  LD_SH4(dst, 64, src0, src1, src2, src3);
+  LD_SH4(dst + 8, 64, src4, src5, src6, src7);
+
+  BUTTERFLY_8(src0, src2, src4, src6, src7, src5, src3, src1, tmp0, tmp2, tmp4,
+              tmp6, tmp7, tmp5, tmp3, tmp1);
+  SRA_4V(tmp0, tmp1, tmp2, tmp3, 1);
+  SRA_4V(tmp4, tmp5, tmp6, tmp7, 1);
+  BUTTERFLY_8(tmp0, tmp1, tmp4, tmp5, tmp7, tmp6, tmp3, tmp2, src0, src1, src4,
+              src5, src7, src6, src3, src2);
+
+  ST_SH4(src0, src1, src2, src3, dst, 64);
+  ST_SH4(src4, src5, src6, src7, dst + 8, 64);
+  dst += 16;
+
+  LD_SH4(dst, 64, src0, src1, src2, src3);
+  LD_SH4(dst + 8, 64, src4, src5, src6, src7);
+
+  BUTTERFLY_8(src0, src2, src4, src6, src7, src5, src3, src1, tmp0, tmp2, tmp4,
+              tmp6, tmp7, tmp5, tmp3, tmp1);
+  SRA_4V(tmp0, tmp1, tmp2, tmp3, 1);
+  SRA_4V(tmp4, tmp5, tmp6, tmp7, 1);
+  BUTTERFLY_8(tmp0, tmp1, tmp4, tmp5, tmp7, tmp6, tmp3, tmp2, src0, src1, src4,
+              src5, src7, src6, src3, src2);
+
+  ST_SH4(src0, src1, src2, src3, dst, 64);
+  ST_SH4(src4, src5, src6, src7, dst + 8, 64);
+  dst += 16;
+
+  LD_SH4(dst, 64, src0, src1, src2, src3);
+  LD_SH4(dst + 8, 64, src4, src5, src6, src7);
+
+  BUTTERFLY_8(src0, src2, src4, src6, src7, src5, src3, src1, tmp0, tmp2, tmp4,
+              tmp6, tmp7, tmp5, tmp3, tmp1);
+  SRA_4V(tmp0, tmp1, tmp2, tmp3, 1);
+  SRA_4V(tmp4, tmp5, tmp6, tmp7, 1);
+  BUTTERFLY_8(tmp0, tmp1, tmp4, tmp5, tmp7, tmp6, tmp3, tmp2, src0, src1, src4,
+              src5, src7, src6, src3, src2);
+
+  ST_SH4(src0, src1, src2, src3, dst, 64);
+  ST_SH4(src4, src5, src6, src7, dst + 8, 64);
+}
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index ee1b29279..4c6984fc7 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -886,10 +886,10 @@ if (vpx_config("CONFIG_VP9_ENCODER") eq "yes") {
   specialize qw/vpx_minmax_8x8 sse2 neon/;
 
   add_proto qw/void vpx_hadamard_8x8/, "const int16_t *src_diff, int src_stride, int16_t *coeff";
-  specialize qw/vpx_hadamard_8x8 sse2 neon/, "$ssse3_x86_64";
+  specialize qw/vpx_hadamard_8x8 sse2 neon msa/, "$ssse3_x86_64";
 
   add_proto qw/void vpx_hadamard_16x16/, "const int16_t *src_diff, int src_stride, int16_t *coeff";
-  specialize qw/vpx_hadamard_16x16 sse2 neon/;
+  specialize qw/vpx_hadamard_16x16 sse2 neon msa/;
 
   add_proto qw/int vpx_satd/, "const int16_t *coeff, int length";
   specialize qw/vpx_satd sse2 neon/;