if no performance drop, we may write as new format At 2015-03-10 07:00:08,dave <[email protected]> wrote:
On 03/09/2015 03:20 PM, chen wrote: a little improve need integrate lea r0, [r0 + r1] when I replace it by "add r0, r1", I got ~3% improve on my Haswell PC This makes no difference on my old machine. At 2015-03-10 05:24:08,[email protected] wrote: ># HG changeset patch ># User David T Yuen <[email protected]> ># 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) > ;--------------------------------------------------------------------------------------------- >_______________________________________________ >x265-devel mailing list >[email protected] >https://mailman.videolan.org/listinfo/x265-devel _______________________________________________ x265-devel mailing list [email protected]https://mailman.videolan.org/listinfo/x265-devel
_______________________________________________ x265-devel mailing list [email protected] https://mailman.videolan.org/listinfo/x265-devel
