[x265] [PATCH] asm: avx2 assembly code for 8bpp transpose16x16 module

murugan at multicorewareinc.com murugan at multicorewareinc.com
Thu Oct 9 13:50:18 CEST 2014


# HG changeset patch
# User Murugan Vairavel <murugan at multicorewareinc.com>
# Date 1412855249 -19800
#      Thu Oct 09 17:17:29 2014 +0530
# Node ID 0c36fca591aefdf1df620126e3e7d2ba327609cb
# Parent  85203bb459124dc5eb4bc929450c655b196aeb0e
asm: avx2 assembly code for 8bpp transpose16x16 module

diff -r 85203bb45912 -r 0c36fca591ae source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Oct 09 15:59:58 2014 +0530
+++ b/source/common/x86/asm-primitives.cpp	Thu Oct 09 17:17:29 2014 +0530
@@ -1792,6 +1792,7 @@
         p.idct[IDCT_8x8] = x265_idct8_avx2;
         p.idct[IDCT_16x16] = x265_idct16_avx2;
         p.idct[IDCT_32x32] = x265_idct32_avx2;
+        p.transpose[BLOCK_16x16] = x265_transpose16_avx2;
         p.transpose[BLOCK_32x32] = x265_transpose32_avx2;
 #endif
     }
diff -r 85203bb45912 -r 0c36fca591ae source/common/x86/pixel-util.h
--- a/source/common/x86/pixel-util.h	Thu Oct 09 15:59:58 2014 +0530
+++ b/source/common/x86/pixel-util.h	Thu Oct 09 17:17:29 2014 +0530
@@ -44,6 +44,7 @@
 void x265_transpose32_sse2(pixel *dest, pixel *src, intptr_t stride);
 void x265_transpose64_sse2(pixel *dest, pixel *src, intptr_t stride);
 
+void x265_transpose16_avx2(pixel *dest, pixel *src, intptr_t stride);
 void x265_transpose32_avx2(pixel *dest, pixel *src, intptr_t stride);
 
 uint32_t x265_quant_sse4(int32_t *coef, int32_t *quantCoeff, int32_t *deltaU, int16_t *qCoef, int qBits, int add, int numCoeff);
diff -r 85203bb45912 -r 0c36fca591ae source/common/x86/pixel-util8.asm
--- a/source/common/x86/pixel-util8.asm	Thu Oct 09 15:59:58 2014 +0530
+++ b/source/common/x86/pixel-util8.asm	Thu Oct 09 17:17:29 2014 +0530
@@ -1632,8 +1632,105 @@
     lea    r0,    [r6 + 8 * r5 + 16]
     mov    r3,    r0
     call   transpose8_internal
-
+    RET
 %else
+%if ARCH_X86_64 == 1
+INIT_YMM avx2
+cglobal transpose16, 3, 5, 16
+    lea          r3, [r2 * 3]
+    lea          r4, [r1 + 8 * r2]
+
+    movu         xm0, [r1]
+    movu         xm1, [r1 + r2]
+    movu         xm2, [r1 + 2 * r2]
+    movu         xm3, [r1 + r3]
+    vinserti128  m0,  m0, [r4], 1
+    vinserti128  m1,  m1, [r4 + r2], 1
+    vinserti128  m2,  m2, [r4 + 2 * r2], 1
+    vinserti128  m3,  m3, [r4 + r3], 1
+    lea          r1,  [r1 + 4 * r2]
+    lea          r4,  [r4 + 4 * r2]
+
+    movu         xm4, [r1]
+    movu         xm5, [r1 + r2]
+    movu         xm6, [r1 + 2 * r2]
+    movu         xm7, [r1 + r3]
+    vinserti128  m4,  m4, [r4], 1
+    vinserti128  m5,  m5, [r4 + r2], 1
+    vinserti128  m6,  m6, [r4 + 2 * r2], 1
+    vinserti128  m7,  m7, [r4 + r3], 1
+
+    punpcklbw    m8, m0, m1                 ;[1 - 8 ; 1 - 8 ][1 2  9 10]
+    punpckhbw    m0, m1                     ;[9 - 16; 9 - 16][1 2  9 10]
+
+    punpcklbw    m1, m2, m3                 ;[1 - 8 ; 1 - 8 ][3 4 11 12]
+    punpckhbw    m2, m3                     ;[9 - 16; 9 - 16][3 4 11 12]
+
+    punpcklbw    m3, m4, m5                 ;[1 - 8 ; 1 - 8 ][5 6 13 14]
+    punpckhbw    m4, m5                     ;[9 - 16; 9 - 16][5 6 13 14]
+
+    punpcklbw    m5, m6, m7                 ;[1 - 8 ; 1 - 8 ][7 8 15 16]
+    punpckhbw    m6, m7                     ;[9 - 16; 9 - 16][7 8 15 16]
+
+    punpcklwd    m7, m8, m1                 ;[1 - 4 ; 1 - 4][1 2 3 4 9 10 11 12]
+    punpckhwd    m8, m1                     ;[5 - 8 ; 5 - 8][1 2 3 4 9 10 11 12]
+
+    punpcklwd    m1, m3, m5                 ;[1 - 4 ; 1 - 4][5 6 7 8 13 14 15 16]
+    punpckhwd    m3, m5                     ;[5 - 8 ; 5 - 8][5 6 7 8 13 14 15 16]
+
+    punpcklwd    m5, m0, m2                 ;[9 - 12; 9 -  12][1 2 3 4 9 10 11 12]
+    punpckhwd    m0, m2                     ;[13- 16; 13 - 16][1 2 3 4 9 10 11 12]
+
+    punpcklwd    m2, m4, m6                 ;[9 - 12; 9 -  12][5 6 7 8 13 14 15 16]
+    punpckhwd    m4, m6                     ;[13- 16; 13 - 16][5 6 7 8 13 14 15 16]
+
+    punpckldq    m6, m7, m1                 ;[1 - 2 ; 1 - 2][1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]
+    punpckhdq    m7, m1                     ;[3 - 4 ; 3 - 4][1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]
+
+    punpckldq    m1, m8, m3                 ;[5 - 6 ; 5 - 6][1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]
+    punpckhdq    m8, m3                     ;[7 - 8 ; 7 - 8][1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]
+
+    punpckldq    m3, m5, m2                 ;[9 - 10; 9 -  10][1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]
+    punpckhdq    m5, m2                     ;[11- 12; 11 - 12][1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]
+
+    punpckldq    m2, m0, m4                 ;[13- 14; 13 - 14][1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]
+    punpckhdq    m0, m4                     ;[15- 16; 15 - 16][1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]
+
+    vpermq       m6, m6, 0xD8
+    vpermq       m7, m7, 0xD8
+    vpermq       m1, m1, 0xD8
+    vpermq       m8, m8, 0xD8
+    vpermq       m3, m3, 0xD8
+    vpermq       m5, m5, 0xD8
+    vpermq       m2, m2, 0xD8
+    vpermq       m0, m0, 0xD8
+
+    movu         [r0 + 0 * 16], xm6
+    vextracti128 [r0 + 1 * 16], m6, 1
+
+    movu         [r0 + 2 * 16], xm7
+    vextracti128 [r0 + 3 * 16], m7, 1
+
+    movu         [r0 + 4 * 16], xm1
+    vextracti128 [r0 + 5 * 16], m1, 1
+
+    movu         [r0 + 6 * 16], xm8
+    vextracti128 [r0 + 7 * 16], m8, 1
+
+    movu         [r0 + 8 * 16], xm3
+    vextracti128 [r0 + 9 * 16], m3, 1
+
+    movu         [r0 + 10 * 16], xm5
+    vextracti128 [r0 + 11 * 16], m5, 1
+
+    movu         [r0 + 12 * 16], xm2
+    vextracti128 [r0 + 13 * 16], m2, 1
+
+    movu         [r0 + 14 * 16], xm0
+    vextracti128 [r0 + 15 * 16], m0, 1
+    RET
+%endif
+INIT_XMM sse2
 cglobal transpose16, 3, 5, 8, dest, src, stride
     mov    r3,    r0
     mov    r4,    r1
@@ -1647,8 +1744,8 @@
     lea    r1,    [r1 + 2 * r2]
     lea    r0,    [r3 + 8 * 16 + 8]
     TRANSPOSE_8x8 16
+    RET
 %endif
-    RET
 
 cglobal transpose16_internal
     TRANSPOSE_8x8 r6
@@ -1761,6 +1858,7 @@
     mov    r5,    r0
     call   transpose16_internal
     RET
+
 %if ARCH_X86_64 == 1
 INIT_YMM avx2
 cglobal transpose32, 3, 5, 16


More information about the x265-devel mailing list