SSSE3 convolution optimization

Optimizing all SSSE3 assembly for convolution:
1. vp9_filter_block1d4_h8_sse2
2. vp9_filter_block1d8_h8_sse2
3. vp9_filter_block1d16_h8_sse2
4. vp9_filter_block1d4_v8_sse2
5. vp9_filter_block1d8_v8_sse2
6. vp9_filter_block1d16_v8_sse2
my optimization include:
-processing 2x8 elements in one 128 bit register instead of processing
8 elements in one 128 bit register.
-removing unecessary loads.
This optimization gives between 2.4% user level gain for 480p input
and 1.6% user level gain for 720p.
This Optimization done only for 64bit.

Change-Id: Icb586dc0c938b56699864fcee6c52fd43b36b969

1 files changed, 93 insertions, 7 deletions

diff --git a/vp9/common/x86/vp9_asm_stubs.c b/vp9/common/x86/vp9_asm_stubs.c
index 106e6d426..b4ab4c306 100644
--- a/vp9/common/x86/vp9_asm_stubs.c
+++ b/vp9/common/x86/vp9_asm_stubs.c
@@ -45,20 +45,105 @@ typedef void filter8_1dfunction (
   const short *filter
 );
 
-#if HAVE_SSSE3
-filter8_1dfunction vp9_filter_block1d16_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d16_h8_ssse3;
-filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
-filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
+#if (HAVE_SSSE3)
 filter8_1dfunction vp9_filter_block1d16_v8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d16_h8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d8_v8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d8_h8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d4_v8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d4_h8_avg_ssse3;
+#if (ARCH_X86_64)
+filter8_1dfunction vp9_filter_block1d16_v8_intrin_ssse3;
+filter8_1dfunction vp9_filter_block1d16_h8_intrin_ssse3;
+filter8_1dfunction vp9_filter_block1d8_v8_intrin_ssse3;
+filter8_1dfunction vp9_filter_block1d8_h8_intrin_ssse3;
+filter8_1dfunction vp9_filter_block1d4_v8_intrin_ssse3;
+filter8_1dfunction vp9_filter_block1d4_h8_intrin_ssse3;
+
+void vp9_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+                               uint8_t *dst, ptrdiff_t dst_stride,
+                               const int16_t *filter_x, int x_step_q4,
+                               const int16_t *filter_y, int y_step_q4,
+                               int w, int h) {
+  /* Ensure the filter can be compressed to int16_t. */
+  if (x_step_q4 == 16 && filter_x[3] != 128) {
+    while (w >= 16) {
+      vp9_filter_block1d16_h8_intrin_ssse3(src, src_stride,
+                                    dst, dst_stride,
+                                    h, filter_x);
+      src += 16;
+      dst += 16;
+      w -= 16;
+    }
+    while (w >= 8) {
+      vp9_filter_block1d8_h8_intrin_ssse3(src, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_x);
+      src += 8;
+      dst += 8;
+      w -= 8;
+    }
+    while (w >= 4) {
+      vp9_filter_block1d4_h8_intrin_ssse3(src, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_x);
+      src += 4;
+      dst += 4;
+      w -= 4;
+    }
+  }
+  if (w) {
+    vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride,
+                          filter_x, x_step_q4, filter_y, y_step_q4,
+                          w, h);
+  }
+}
 
+void vp9_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+                              uint8_t *dst, ptrdiff_t dst_stride,
+                              const int16_t *filter_x, int x_step_q4,
+                              const int16_t *filter_y, int y_step_q4,
+                              int w, int h) {
+  if (y_step_q4 == 16 && filter_y[3] != 128) {
+    while (w >= 16) {
+      vp9_filter_block1d16_v8_intrin_ssse3(src - src_stride * 3, src_stride,
+                                    dst, dst_stride,
+                                    h, filter_y);
+      src += 16;
+      dst += 16;
+      w -= 16;
+    }
+    while (w >= 8) {
+      vp9_filter_block1d8_v8_intrin_ssse3(src - src_stride * 3, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_y);
+      src += 8;
+      dst += 8;
+      w -= 8;
+    }
+    while (w >= 4) {
+      vp9_filter_block1d4_v8_intrin_ssse3(src - src_stride * 3, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_y);
+      src += 4;
+      dst += 4;
+      w -= 4;
+    }
+  }
+  if (w) {
+    vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
+                         filter_x, x_step_q4, filter_y, y_step_q4,
+                         w, h);
+  }
+}
+
+#else
+filter8_1dfunction vp9_filter_block1d16_v8_ssse3;
+filter8_1dfunction vp9_filter_block1d16_h8_ssse3;
+filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
+filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
+filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
+filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
 void vp9_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
                                uint8_t *dst, ptrdiff_t dst_stride,
                                const int16_t *filter_x, int x_step_q4,
@@ -135,6 +220,7 @@ void vp9_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
                          w, h);
   }
 }
+#endif
 
 void vp9_convolve8_avg_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
                                uint8_t *dst, ptrdiff_t dst_stride,