Enable hbd_build to use SSSE3optimized functions

This commit changes the SSSE3 assembly functions for idct32x32 to
support highbitdepth build.

On test clip fdJc1_IBKJA.248.webm, this cuts the speed difference
between hbd and lbd build from between 3-4% to 1-2%.

Change-Id: Ic3390e0113bc1ca5bba8ec80d1795ad31b484fca

2 files changed, 77 insertions, 5 deletions

diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index ff994e30e..d3c497105 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -803,14 +803,15 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
     specialize qw/vpx_idct16x16_1_add sse2/;
 
     add_proto qw/void vpx_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
-    specialize qw/vpx_idct32x32_1024_add sse2/;
+    specialize qw/vpx_idct32x32_1024_add sse2/, "$ssse3_x86_64_x86inc";
 
     add_proto qw/void vpx_idct32x32_135_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
-    specialize qw/vpx_idct32x32_135_add sse2/;
+    specialize qw/vpx_idct32x32_135_add sse2/, "$ssse3_x86_64_x86inc";
+    # Need to add 135 eob idct32x32 implementations.
     $vpx_idct32x32_135_add_sse2=vpx_idct32x32_1024_add_sse2;
 
     add_proto qw/void vpx_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
-    specialize qw/vpx_idct32x32_34_add sse2/;
+    specialize qw/vpx_idct32x32_34_add sse2/, "$ssse3_x86_64_x86inc";
 
     add_proto qw/void vpx_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
     specialize qw/vpx_idct32x32_1_add sse2/;
diff --git a/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm b/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm
index 80a330b7b..7cb9a4192 100644
--- a/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm
+++ b/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm
@@ -765,6 +765,24 @@ idct32x32_34:
   lea             r4, [rsp + transposed_in]
 
 idct32x32_34_transpose:
+%if CONFIG_VP9_HIGHBITDEPTH
+  mova            m0, [r3 +       0]
+  packssdw        m0, [r3 +      16]
+  mova            m1, [r3 + 32 *  4]
+  packssdw        m1, [r3 + 32 *  4 + 16]
+  mova            m2, [r3 + 32 *  8]
+  packssdw        m2, [r3 + 32 *  8 + 16]
+  mova            m3, [r3 + 32 * 12]
+  packssdw        m3, [r3 + 32 * 12 + 16]
+  mova            m4, [r3 + 32 * 16]
+  packssdw        m4, [r3 + 32 * 16 + 16]
+  mova            m5, [r3 + 32 * 20]
+  packssdw        m5, [r3 + 32 * 20 + 16]
+  mova            m6, [r3 + 32 * 24]
+  packssdw        m6, [r3 + 32 * 24 + 16]
+  mova            m7, [r3 + 32 * 28]
+  packssdw        m7, [r3 + 32 * 28 + 16]
+%else
   mova            m0, [r3 +       0]
   mova            m1, [r3 + 16 *  4]
   mova            m2, [r3 + 16 *  8]
@@ -773,6 +791,7 @@ idct32x32_34_transpose:
   mova            m5, [r3 + 16 * 20]
   mova            m6, [r3 + 16 * 24]
   mova            m7, [r3 + 16 * 28]
+%endif
 
   TRANSPOSE8X8  0, 1, 2, 3, 4, 5, 6, 7, 9
 
@@ -1176,6 +1195,24 @@ idct32x32_135:
   mov             r7, 2
 
 idct32x32_135_transpose:
+%if CONFIG_VP9_HIGHBITDEPTH
+  mova            m0, [r3 +       0]
+  packssdw        m0, [r3 +      16]
+  mova            m1, [r3 + 32 *  4]
+  packssdw        m1, [r3 + 32 *  4 + 16]
+  mova            m2, [r3 + 32 *  8]
+  packssdw        m2, [r3 + 32 *  8 + 16]
+  mova            m3, [r3 + 32 * 12]
+  packssdw        m3, [r3 + 32 * 12 + 16]
+  mova            m4, [r3 + 32 * 16]
+  packssdw        m4, [r3 + 32 * 16 + 16]
+  mova            m5, [r3 + 32 * 20]
+  packssdw        m5, [r3 + 32 * 20 + 16]
+  mova            m6, [r3 + 32 * 24]
+  packssdw        m6, [r3 + 32 * 24 + 16]
+  mova            m7, [r3 + 32 * 28]
+  packssdw        m7, [r3 + 32 * 28 + 16]
+%else
   mova            m0, [r3 +       0]
   mova            m1, [r3 + 16 *  4]
   mova            m2, [r3 + 16 *  8]
@@ -1184,7 +1221,7 @@ idct32x32_135_transpose:
   mova            m5, [r3 + 16 * 20]
   mova            m6, [r3 + 16 * 24]
   mova            m7, [r3 + 16 * 28]
-
+%endif
   TRANSPOSE8X8  0, 1, 2, 3, 4, 5, 6, 7, 9
 
   mova [r4 +      0], m0
@@ -1196,14 +1233,22 @@ idct32x32_135_transpose:
   mova [r4 + 16 * 6], m6
   mova [r4 + 16 * 7], m7
 
+%if CONFIG_VP9_HIGHBITDEPTH
+  add             r3, 32
+%else
   add             r3, 16
+%endif
   add             r4, 16 * 8
   dec             r7
   jne idct32x32_135_transpose
 
   IDCT32X32_135 16*0, 16*32, 16*64, 16*96
   lea            stp, [stp + 16 * 8]
+%if CONFIG_VP9_HIGHBITDEPTH
+  lea         inputq, [inputq + 32 * 32]
+%else
   lea         inputq, [inputq + 16 * 32]
+%endif
   dec             r6
   jnz idct32x32_135
 
@@ -1614,6 +1659,24 @@ idct32x32_1024:
   mov             r7, 4
 
 idct32x32_1024_transpose:
+%if CONFIG_VP9_HIGHBITDEPTH
+  mova            m0, [r3 +       0]
+  packssdw        m0, [r3 +      16]
+  mova            m1, [r3 + 32 *  4]
+  packssdw        m1, [r3 + 32 *  4 + 16]
+  mova            m2, [r3 + 32 *  8]
+  packssdw        m2, [r3 + 32 *  8 + 16]
+  mova            m3, [r3 + 32 * 12]
+  packssdw        m3, [r3 + 32 * 12 + 16]
+  mova            m4, [r3 + 32 * 16]
+  packssdw        m4, [r3 + 32 * 16 + 16]
+  mova            m5, [r3 + 32 * 20]
+  packssdw        m5, [r3 + 32 * 20 + 16]
+  mova            m6, [r3 + 32 * 24]
+  packssdw        m6, [r3 + 32 * 24 + 16]
+  mova            m7, [r3 + 32 * 28]
+  packssdw        m7, [r3 + 32 * 28 + 16]
+%else
   mova            m0, [r3 +       0]
   mova            m1, [r3 + 16 *  4]
   mova            m2, [r3 + 16 *  8]
@@ -1622,6 +1685,7 @@ idct32x32_1024_transpose:
   mova            m5, [r3 + 16 * 20]
   mova            m6, [r3 + 16 * 24]
   mova            m7, [r3 + 16 * 28]
+%endif
 
   TRANSPOSE8X8  0, 1, 2, 3, 4, 5, 6, 7, 9
 
@@ -1633,8 +1697,11 @@ idct32x32_1024_transpose:
   mova [r4 + 16 * 5], m5
   mova [r4 + 16 * 6], m6
   mova [r4 + 16 * 7], m7
-
+%if CONFIG_VP9_HIGHBITDEPTH
+  add             r3, 32
+%else
   add             r3, 16
+%endif
   add             r4, 16 * 8
   dec             r7
   jne idct32x32_1024_transpose
@@ -1642,7 +1709,11 @@ idct32x32_1024_transpose:
   IDCT32X32_1024 16*0, 16*32, 16*64, 16*96
 
   lea            stp, [stp + 16 * 8]
+%if CONFIG_VP9_HIGHBITDEPTH
+  lea         inputq, [inputq + 32 * 32]
+%else
   lea         inputq, [inputq + 16 * 32]
+%endif
   dec             r6
   jnz idct32x32_1024