Add an SSE2 version of vp9_iwht4x4_16_add.

80% fewer cycles than C

Change-Id: I841bde1e268ddd33ae2ee75eee94737a400e2cde

test/fdct4x4_test.cc

@@ -494,6 +494,14 @@ INSTANTIATE_TEST_CASE_P(
         make_tuple(&vp9_fwht4x4_mmx, &vp9_iwht4x4_16_add_c, 0, VPX_BITS_8)));
 #endif
 
+#if CONFIG_USE_X86INC && HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && \
+    !CONFIG_EMULATE_HARDWARE
+INSTANTIATE_TEST_CASE_P(
+    SSE2, Trans4x4WHT,
+    ::testing::Values(
+        make_tuple(&vp9_fwht4x4_c, &vp9_iwht4x4_16_add_sse2, 0, VPX_BITS_8)));
+#endif
+
 #if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
 INSTANTIATE_TEST_CASE_P(
     SSE2, Trans4x4DCT,

vp9/common/vp9_rtcd_defs.pl

@@ -475,7 +475,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
     specialize qw/vp9_iwht4x4_1_add msa/;
 
     add_proto qw/void vp9_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
-    specialize qw/vp9_iwht4x4_16_add msa/;
+    specialize qw/vp9_iwht4x4_16_add msa/, "$sse2_x86inc";
   }
 }
 

vp9/common/x86/vp9_idct_sse2.asm

+;
+;  Copyright (c) 2015 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 "third_party/x86inc/x86inc.asm"
+
+SECTION .text
+
+%macro TRANSFORM_COLS 0
+  paddw           m0,        m2
+  psubw           m3,        m1
+  psubw           m4,        m0,        m3
+  psraw           m4,        1
+  psubw           m5,        m4,        m1 ;b1
+  psubw           m4,        m2 ;c1
+  psubw           m0,        m5
+  paddw           m3,        m4
+                                ; m0 a0
+  SWAP            1,         4  ; m1 c1
+  SWAP            2,         3  ; m2 d1
+  SWAP            3,         5  ; m3 b1
+%endmacro
+
+%macro TRANSPOSE_4X4 0
+  SWAP            4, 0
+  SWAP            5, 2
+  punpcklwd       m4,        m1
+  pshufd          m0,        m4, 0x0e
+  punpcklwd       m5,        m3
+  pshufd          m2,        m5, 0x0e
+  SWAP            1, 4
+  SWAP            3, 0
+  punpckldq       m1,        m5
+  pshufd          m4,        m1, 0x0e
+  punpckldq       m3,        m2
+  pshufd          m0,        m3, 0x0e
+  SWAP            2, 3, 0, 1, 4
+%endmacro
+
+; transposes a 4x4 int16 matrix in xmm0 and xmm1 to the bottom half of xmm0-xmm3
+%macro TRANSPOSE_4X4_WIDE 0
+  mova            m3, m0
+  punpcklwd       m0, m1
+  punpckhwd       m3, m1
+  mova            m2, m0
+  punpcklwd       m0, m3
+  punpckhwd       m2, m3
+  pshufd          m1, m0, 0x0e
+  pshufd          m3, m2, 0x0e
+%endmacro
+
+%macro ADD_STORE_4P_2X 5  ; src1, src2, tmp1, tmp2, zero
+  movq            m%3,       [outputq]
+  movq            m%4,       [outputq + strideq]
+  punpcklbw       m%3,       m%5
+  punpcklbw       m%4,       m%5
+  paddw           m%1,       m%3
+  paddw           m%2,       m%4
+  packuswb        m%1,       m%5
+  packuswb        m%2,       m%5
+  movd            [outputq], m%1
+  movd            [outputq + strideq], m%2
+%endmacro
+
+INIT_XMM sse2
+cglobal iwht4x4_16_add, 3, 4, 7, input, output, stride
+  mova            m0,        [inputq +  0] ;a1
+  mova            m1,        [inputq + 16] ;c1
+
+  psraw           m0,        2
+  psraw           m1,        2
+
+  TRANSPOSE_4X4_WIDE
+  TRANSFORM_COLS
+  TRANSPOSE_4X4
+  TRANSFORM_COLS
+
+  pxor            m4, m4
+  ADD_STORE_4P_2X  0, 1, 5, 6, 4
+  lea             outputq, [outputq + 2 * strideq]
+  ADD_STORE_4P_2X  2, 3, 5, 6, 4
+
+  RET

vp9/vp9_common.mk

@@ -75,6 +75,7 @@ VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.h
 VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.c
 VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_mfqe.h
 VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_mfqe.c
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_sse2.asm
 VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_loopfilter_mmx.asm
 VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_subpixel_8t_sse2.asm
 VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_subpixel_bilinear_sse2.asm