[x265] [PATCH 1 of 2] asm:intra pred planar32 sse2

dtyx265 at gmail.com dtyx265 at gmail.com
Mon Mar 9 22:24:08 CET 2015


# HG changeset patch
# User David T Yuen <dtyx265 at gmail.com>
# Date 1425934257 25200
# Node ID ef383507c21ca704b58e759e440f4ae2e177c499
# Parent  c16a875d913fffbad405f7d5af51700b1a8990bb
asm:intra pred planar32 sse2

This replaces c code for systems using ssse3 to sse2 processors
The code is backported from intrapred planar32 sse4

There are essentially are two versions here.
One for x86_64 and one for x86_32.  It would have been too ugly
to conditionally code the differences in a single primitive.

64-bit

./test/TestBench --testbench intrapred | grep intra_planar_32x32
intra_planar_32x32	13.12x 	 9254.99  	 121439.99

32-bit

./test/TestBench --testbench intrapred | grep intra_planar_32x32
intra_planar_32x32	10.20x 	 9720.04  	 99170.14

diff -r c16a875d913f -r ef383507c21c source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Mar 05 16:01:49 2015 -0800
+++ b/source/common/x86/asm-primitives.cpp	Mon Mar 09 13:50:57 2015 -0700
@@ -1221,6 +1221,7 @@
         p.cu[BLOCK_4x4].intra_pred[PLANAR_IDX] = x265_intra_pred_planar4_sse2;
         p.cu[BLOCK_8x8].intra_pred[PLANAR_IDX] = x265_intra_pred_planar8_sse2;
         p.cu[BLOCK_16x16].intra_pred[PLANAR_IDX] = x265_intra_pred_planar16_sse2;
+        p.cu[BLOCK_32x32].intra_pred[PLANAR_IDX] = x265_intra_pred_planar32_sse2;
 
         p.cu[BLOCK_4x4].calcresidual = x265_getResidual4_sse2;
         p.cu[BLOCK_8x8].calcresidual = x265_getResidual8_sse2;
diff -r c16a875d913f -r ef383507c21c source/common/x86/intrapred.h
--- a/source/common/x86/intrapred.h	Thu Mar 05 16:01:49 2015 -0800
+++ b/source/common/x86/intrapred.h	Mon Mar 09 13:50:57 2015 -0700
@@ -38,6 +38,7 @@
 void x265_intra_pred_planar4_sse2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
 void x265_intra_pred_planar8_sse2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
 void x265_intra_pred_planar16_sse2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
+void x265_intra_pred_planar32_sse2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
 void x265_intra_pred_planar4_sse4(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
 void x265_intra_pred_planar8_sse4(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
 void x265_intra_pred_planar16_sse4(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
diff -r c16a875d913f -r ef383507c21c source/common/x86/intrapred8.asm
--- a/source/common/x86/intrapred8.asm	Thu Mar 05 16:01:49 2015 -0800
+++ b/source/common/x86/intrapred8.asm	Mon Mar 09 13:50:57 2015 -0700
@@ -734,6 +734,270 @@
     INTRA_PRED_PLANAR_16 15
     RET
 
+;---------------------------------------------------------------------------------------
+; void intra_pred_planar(pixel* dst, intptr_t dstStride, pixel*srcPix, int, int filter)
+;---------------------------------------------------------------------------------------
+INIT_XMM sse2
+%macro PROCESS 1
+    pmullw          m5, %1, [pw_planar32_L]
+    pmullw          m6, %1, [pw_planar32_H]
+    paddw           m5, m0
+    paddw           m6, m1
+    psraw           m5, 6
+    psraw           m6, 6
+    packuswb        m5, m6
+    movu            [r0], m5
+
+    pmullw          m5, %1, [pw_planar16_0]
+    pmullw          %1, [pw_planar8_0]
+    paddw           m5, m2
+    paddw           %1, m3
+    psraw           m5, 6
+    psraw           %1, 6
+    packuswb        m5, %1
+    movu            [r0 + 16], m5
+%endmacro
+
+%if ARCH_X86_64 == 1
+cglobal intra_pred_planar32, 3,3,14
+    movd            m3, [r2 + 33]               ; topRight   = above[32]
+
+    pxor            m7, m7
+    pand            m3, [pw_00ff]
+    pshuflw         m3, m3, 0x00
+    pshufd          m3, m3, 0x44
+
+    pmullw          m0, m3, [multiL]            ; (x + 1) * topRight
+    pmullw          m1, m3, [multiH]            ; (x + 1) * topRight
+    pmullw          m2, m3, [multiH2]           ; (x + 1) * topRight
+    pmullw          m3, [multiH3]               ; (x + 1) * topRight
+
+    movd            m11, [r2 + 97]               ; bottomLeft = left[32]
+    pand            m11, [pw_00ff]
+    pshuflw         m11, m11, 0x00
+    pshufd          m11, m11, 0x44
+    mova            m5,  m11
+    paddw           m5,  [pw_32]
+
+    paddw           m0, m5
+    paddw           m1, m5
+    paddw           m2, m5
+    paddw           m3, m5
+    mova            m8, m11
+    mova            m9, m11
+    mova            m10, m11
+
+    movh            m4, [r2 + 1]
+    punpcklbw       m4, m7
+    psubw           m8, m4
+    pmullw          m4, [pw_planar32_1]
+    paddw           m0, m4
+
+    movh            m4, [r2 + 9]
+    punpcklbw       m4, m7
+    psubw           m9, m4
+    pmullw          m4, [pw_planar32_1]
+    paddw           m1, m4
+
+    movh            m4, [r2 + 17]
+    punpcklbw       m4, m7
+    psubw           m10, m4
+    pmullw          m4, [pw_planar32_1]
+    paddw           m2, m4
+
+    movh            m4, [r2 + 25]
+    punpcklbw       m4, m7
+    psubw           m11, m4
+    pmullw          m4, [pw_planar32_1]
+    paddw           m3, m4
+
+%macro INCREMENT 0
+    paddw           m2, m10
+    paddw           m3, m11
+    paddw           m0, m8
+    paddw           m1, m9
+    lea             r0, [r0 + r1]
+%endmacro
+
+%assign x 0
+%rep 2
+    movu            m4, [r2 + 65 + x * 16]
+    movu            m7, m4
+    pand            m4, [pw_00ff]
+    psrlw           m7, 8
+    pshuflw         m12, m4, 0x00
+    pshufd          m12, m12, 0x44
+    pshuflw         m13, m7, 0x00
+    pshufd          m13, m13, 0x44
+
+    PROCESS m12
+    INCREMENT
+    PROCESS m13
+    INCREMENT
+
+    pshuflw         m12, m4, 0x55
+    pshufd          m12, m12, 0x44
+    pshuflw         m13, m7, 0x55
+    pshufd          m13, m13, 0x44
+
+    PROCESS m12
+    INCREMENT
+    PROCESS m13
+    INCREMENT
+
+    pshuflw         m12, m4, 0xAA
+    pshufd          m12, m12, 0x44
+    pshuflw         m13, m7, 0xAA
+    pshufd          m13, m13, 0x44
+
+    PROCESS m12
+    INCREMENT
+    PROCESS m13
+    INCREMENT
+
+    pshuflw         m12, m4, 0xFF
+    pshufd          m12, m12, 0x44
+    pshuflw         m13, m7, 0xFF
+    pshufd          m13, m13, 0x44
+
+    PROCESS m12
+    INCREMENT
+    PROCESS m13
+    INCREMENT
+
+    pshufhw         m12, m4, 0x00
+    pshufd          m12, m12, 0xEE
+    pshufhw         m13, m7, 0x00
+    pshufd          m13, m13, 0xEE
+
+    PROCESS m12
+    INCREMENT
+    PROCESS m13
+    INCREMENT
+
+    pshufhw         m12, m4, 0x55
+    pshufd          m12, m12, 0xEE
+    pshufhw         m13, m7, 0x55
+    pshufd          m13, m13, 0xEE
+
+    PROCESS m12
+    INCREMENT
+    PROCESS m13
+    INCREMENT
+
+    pshufhw         m12, m4, 0xAA
+    pshufd          m12, m12, 0xEE
+    pshufhw         m13, m7, 0xAA
+    pshufd          m13, m13, 0xEE
+
+    PROCESS m12
+    INCREMENT
+    PROCESS m13
+    INCREMENT
+
+    pshufhw         m12, m4, 0xFF
+    pshufd          m12, m12, 0xEE
+    pshufhw         m13, m7, 0xFF
+    pshufd          m13, m13, 0xEE
+
+    PROCESS m12
+    INCREMENT
+    PROCESS m13
+
+    %if x < 1
+    INCREMENT
+    %endif
+%assign x x+1
+%endrep
+    RET
+
+%else ;end ARCH_X86_64, start ARCH_X86_32
+cglobal intra_pred_planar32, 3,3,8,0-(4*mmsize)
+    movd            m3, [r2 + 33]               ; topRight   = above[32]
+
+    pxor            m7, m7
+    pand            m3, [pw_00ff]
+    pshuflw         m3, m3, 0x00
+    pshufd          m3, m3, 0x44
+
+    pmullw          m0, m3, [multiL]            ; (x + 1) * topRight
+    pmullw          m1, m3, [multiH]            ; (x + 1) * topRight
+    pmullw          m2, m3, [multiH2]           ; (x + 1) * topRight
+    pmullw          m3, [multiH3]               ; (x + 1) * topRight
+
+    movd            m6, [r2 + 97]               ; bottomLeft = left[32]
+    pand            m6, [pw_00ff]
+    pshuflw         m6, m6, 0x00
+    pshufd          m6, m6, 0x44
+    mova            m5, m6
+    paddw           m5, [pw_32]
+
+    paddw           m0, m5
+    paddw           m1, m5
+    paddw           m2, m5
+    paddw           m3, m5
+
+    movh            m4, [r2 + 1]
+    punpcklbw       m4, m7
+    psubw           m5, m6, m4
+    mova            [rsp + 0 * mmsize], m5
+    pmullw          m4, [pw_planar32_1]
+    paddw           m0, m4
+
+    movh            m4, [r2 + 9]
+    punpcklbw       m4, m7
+    psubw           m5, m6, m4
+    mova            [rsp + 1 * mmsize], m5
+    pmullw          m4, [pw_planar32_1]
+    paddw           m1, m4
+
+    movh            m4, [r2 + 17]
+    punpcklbw       m4, m7
+    psubw           m5, m6, m4
+    mova            [rsp + 2 * mmsize], m5
+    pmullw          m4, [pw_planar32_1]
+    paddw           m2, m4
+
+    movh            m4, [r2 + 25]
+    punpcklbw       m4, m7
+    psubw           m5, m6, m4
+    mova            [rsp + 3 * mmsize], m5
+    pmullw          m4, [pw_planar32_1]
+    paddw           m3, m4
+
+%macro INCREMENT 0
+    paddw           m0, [rsp + 0 * mmsize]
+    paddw           m1, [rsp + 1 * mmsize]
+    paddw           m2, [rsp + 2 * mmsize]
+    paddw           m3, [rsp + 3 * mmsize]
+    lea             r0, [r0 + r1]
+%endmacro
+
+%assign x 0
+%rep 16
+    movd            m4, [r2 + 65 + x * 2]
+    movq            m7, m4
+    pand            m4, [pw_00ff]
+    psrlw           m7, 8
+
+    pshuflw         m4, m4, 0x00
+    pshufd          m4, m4, 0x44
+
+    pshuflw         m7, m7, 0x00
+    pshufd          m7, m7, 0x44
+
+    PROCESS m4
+    INCREMENT
+    PROCESS m7
+    %if x < 15
+    INCREMENT
+    %endif
+%assign x x+1
+%endrep
+    RET
+
+%endif ; end ARCH_X86_32
+
 ;---------------------------------------------------------------------------------------------
 ; void intra_pred_dc(pixel* dst, intptr_t dstStride, pixel *srcPix, int dirMode, int bFilter)
 ;---------------------------------------------------------------------------------------------


More information about the x265-devel mailing list