Replacing vp9_get_mb_ss_sse2 asm implementation with intrinsics.

Change-Id: Ib4f5dd733eb2939b108070a01e83da5d9990bac0

test/variance_test.cc

@@ -35,6 +35,14 @@ using ::std::tr1::make_tuple;
 using ::std::tr1::tuple;
 using libvpx_test::ACMRandom;
 
+static unsigned int mb_ss_ref(const int16_t *src) {
+  unsigned int res = 0;
+  for (int i = 0; i < 256; ++i) {
+    res += src[i] * src[i];
+  }
+  return res;
+}
+
 static unsigned int variance_ref(const uint8_t *ref, const uint8_t *src,
                                  int l2w, int l2h, unsigned int *sse_ptr) {
   int se = 0;
@@ -76,6 +84,50 @@ static unsigned int subpel_variance_ref(const uint8_t *ref, const uint8_t *src,
   return sse - (((int64_t) se * se) >> (l2w + l2h));
 }
 
+typedef unsigned int (*SumOfSquaresFunction)(const int16_t *src);
+
+class SumOfSquaresTest : public ::testing::TestWithParam<SumOfSquaresFunction> {
+ public:
+  SumOfSquaresTest() : func_(GetParam()) {}
+
+  virtual ~SumOfSquaresTest() {
+    libvpx_test::ClearSystemState();
+  }
+
+ protected:
+  void ConstTest();
+  void RefTest();
+
+  SumOfSquaresFunction func_;
+  ACMRandom rnd_;
+};
+
+void SumOfSquaresTest::ConstTest() {
+  int16_t mem[256];
+  unsigned int res;
+  for (int v = 0; v < 256; ++v) {
+    for (int i = 0; i < 256; ++i) {
+      mem[i] = v;
+    }
+    ASM_REGISTER_STATE_CHECK(res = func_(mem));
+    EXPECT_EQ(256u * (v * v), res);
+  }
+}
+
+void SumOfSquaresTest::RefTest() {
+  int16_t mem[256];
+  for (int i = 0; i < 100; ++i) {
+    for (int j = 0; j < 256; ++j) {
+      mem[j] = rnd_.Rand8() - rnd_.Rand8();
+    }
+
+    const unsigned int expected = mb_ss_ref(mem);
+    unsigned int res;
+    ASM_REGISTER_STATE_CHECK(res = func_(mem));
+    EXPECT_EQ(expected, res);
+  }
+}
+
 template<typename VarianceFunctionType>
 class VarianceTest
     : public ::testing::TestWithParam<tuple<int, int, VarianceFunctionType> > {
@@ -362,6 +414,13 @@ INSTANTIATE_TEST_CASE_P(
 namespace vp9 {
 
 #if CONFIG_VP9_ENCODER
+
+TEST_P(SumOfSquaresTest, Const) { ConstTest(); }
+TEST_P(SumOfSquaresTest, Ref) { RefTest(); }
+
+INSTANTIATE_TEST_CASE_P(C, SumOfSquaresTest,
+                        ::testing::Values(vp9_get_mb_ss_c));
+
 typedef VarianceTest<vp9_variance_fn_t> VP9VarianceTest;
 typedef SubpelVarianceTest<vp9_subpixvariance_fn_t> VP9SubpelVarianceTest;
 typedef SubpelVarianceTest<vp9_subp_avg_variance_fn_t> VP9SubpelAvgVarianceTest;
@@ -487,6 +546,10 @@ INSTANTIATE_TEST_CASE_P(
 
 #if HAVE_SSE2
 #if CONFIG_USE_X86INC
+
+INSTANTIATE_TEST_CASE_P(SSE2, SumOfSquaresTest,
+                        ::testing::Values(vp9_get_mb_ss_sse2));
+
 const vp9_variance_fn_t variance4x4_sse2 = vp9_variance4x4_sse2;
 const vp9_variance_fn_t variance4x8_sse2 = vp9_variance4x8_sse2;
 const vp9_variance_fn_t variance8x4_sse2 = vp9_variance8x4_sse2;

vp9/encoder/vp9_variance.c

@@ -103,8 +103,9 @@ static void var_filter_block2d_bil_second_pass(const uint16_t *src_ptr,
 unsigned int vp9_get_mb_ss_c(const int16_t *src_ptr) {
   unsigned int i, sum = 0;
 
-  for (i = 0; i < 256; i++)
+  for (i = 0; i < 256; ++i) {
     sum += src_ptr[i] * src_ptr[i];
+  }
 
   return sum;
 }

vp9/encoder/x86/vp9_variance_impl_sse2.asm

@@ -1,69 +0,0 @@
-;
-;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-;  Use of this source code is governed by a BSD-style license
-;  that can be found in the LICENSE file in the root of the source
-;  tree. An additional intellectual property rights grant can be found
-;  in the file PATENTS.  All contributing project authors may
-;  be found in the AUTHORS file in the root of the source tree.
-;
-
-
-%include "vpx_ports/x86_abi_support.asm"
-
-;unsigned int vp9_get_mb_ss_sse2
-;(
-;    short *src_ptr
-;)
-global sym(vp9_get_mb_ss_sse2) PRIVATE
-sym(vp9_get_mb_ss_sse2):
-    push        rbp
-    mov         rbp, rsp
-    SHADOW_ARGS_TO_STACK 1
-    GET_GOT     rbx
-    push rsi
-    push rdi
-    sub         rsp, 16
-    ; end prolog
-
-
-        mov         rax, arg(0) ;[src_ptr]
-        mov         rcx, 8
-        pxor        xmm4, xmm4
-
-.NEXTROW:
-        movdqa      xmm0, [rax]
-        movdqa      xmm1, [rax+16]
-        movdqa      xmm2, [rax+32]
-        movdqa      xmm3, [rax+48]
-        pmaddwd     xmm0, xmm0
-        pmaddwd     xmm1, xmm1
-        pmaddwd     xmm2, xmm2
-        pmaddwd     xmm3, xmm3
-
-        paddd       xmm0, xmm1
-        paddd       xmm2, xmm3
-        paddd       xmm4, xmm0
-        paddd       xmm4, xmm2
-
-        add         rax, 0x40
-        dec         rcx
-        ja          .NEXTROW
-
-        movdqa      xmm3,xmm4
-        psrldq      xmm4,8
-        paddd       xmm4,xmm3
-        movdqa      xmm3,xmm4
-        psrldq      xmm4,4
-        paddd       xmm4,xmm3
-        movq        rax,xmm4
-
-
-    ; begin epilog
-    add rsp, 16
-    pop rdi
-    pop rsi
-    RESTORE_GOT
-    UNSHADOW_ARGS
-    pop         rbp
-    ret

vp9/encoder/x86/vp9_variance_sse2.c

@@ -19,6 +19,21 @@ typedef unsigned int (*variance_fn_t) (const unsigned char *src, int src_stride,
                                        const unsigned char *ref, int ref_stride,
                                        unsigned int *sse, int *sum);
 
+unsigned int vp9_get_mb_ss_sse2(const int16_t *src) {
+  __m128i vsum = _mm_setzero_si128();
+  int i;
+
+  for (i = 0; i < 32; ++i) {
+    const __m128i v = _mm_loadu_si128((const __m128i *)src);
+    vsum = _mm_add_epi32(vsum, _mm_madd_epi16(v, v));
+    src += 8;
+  }
+
+  vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 8));
+  vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 4));
+  return  _mm_cvtsi128_si32(vsum);
+}
+
 #define READ64(p, stride, i) \
   _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const uint32_t *)(p + i * stride)), \
       _mm_cvtsi32_si128(*(const uint32_t *)(p + (i + 1) * stride)))

vp9/vp9cx.mk

@@ -93,7 +93,6 @@ VP9_CX_SRCS-yes += encoder/vp9_temporal_filter.h
 VP9_CX_SRCS-yes += encoder/vp9_mbgraph.c
 VP9_CX_SRCS-yes += encoder/vp9_mbgraph.h
 
-VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_variance_impl_sse2.asm
 VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_variance_impl_intrin_avx2.c
 VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_sad4d_sse2.asm
 VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_sad4d_intrin_avx2.c