[x265] [PATCH] blockcopy_pp: 32x8, 32x16, 32x24, 32x32, 32x48, 32x64 AVX version of asm code, approx double speedup comapre to SSE

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari
# Date 1411119080 -19800
# Node ID e441b1fe3f7870f4aa066680f7a55e547f13dfb8
# Parent  4680ab4f92b8cc809b1e8dbc927126ec70bcc5c5
blockcopy_pp: 32x8, 32x16, 32x24, 32x32, 32x48, 32x64 AVX version of asm code, approx double speedup comapre to SSE

diff -r 4680ab4f92b8 -r e441b1fe3f78 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Sep 18 18:16:25 2014 +0530
+++ b/source/common/x86/asm-primitives.cpp	Fri Sep 19 15:01:20 2014 +0530
@@ -1709,6 +1709,24 @@
 
         p.ssim_4x4x2_core = x265_pixel_ssim_4x4x2_core_avx;
         p.ssim_end_4 = x265_pixel_ssim_end4_avx;
+
+       /* Will be replaced with macro calls once all the partitions are coded */
+
+        p.chroma[X265_CSP_I420].copy_pp[CHROMA_32x8]  = x265_blockcopy_pp_32x8_avx;
+        p.chroma[X265_CSP_I420].copy_pp[CHROMA_32x16] = x265_blockcopy_pp_32x16_avx;
+        p.chroma[X265_CSP_I420].copy_pp[CHROMA_32x24] = x265_blockcopy_pp_32x24_avx;
+        p.chroma[X265_CSP_I420].copy_pp[CHROMA_32x32] = x265_blockcopy_pp_32x32_avx;
+
+        p.chroma[X265_CSP_I422].copy_pp[CHROMA422_32x16] = x265_blockcopy_pp_32x16_avx;
+        p.chroma[X265_CSP_I422].copy_pp[CHROMA422_32x32] = x265_blockcopy_pp_32x32_avx;
+        p.chroma[X265_CSP_I422].copy_pp[CHROMA422_32x48] = x265_blockcopy_pp_32x48_avx;
+        p.chroma[X265_CSP_I422].copy_pp[CHROMA422_32x64] = x265_blockcopy_pp_32x64_avx;
+
+        p.luma_copy_pp[LUMA_32x8]   = x265_blockcopy_pp_32x8_avx;
+        p.luma_copy_pp[LUMA_32x16]  = x265_blockcopy_pp_32x16_avx;
+        p.luma_copy_pp[LUMA_32x24]  = x265_blockcopy_pp_32x24_avx;
+        p.luma_copy_pp[LUMA_32x32]  = x265_blockcopy_pp_32x32_avx;
+        p.luma_copy_pp[LUMA_32x64]  = x265_blockcopy_pp_32x64_avx;
     }
     if (cpuMask & X265_CPU_XOP)
     {
diff -r 4680ab4f92b8 -r e441b1fe3f78 source/common/x86/blockcopy8.asm
--- a/source/common/x86/blockcopy8.asm	Thu Sep 18 18:16:25 2014 +0530
+++ b/source/common/x86/blockcopy8.asm	Fri Sep 19 15:01:20 2014 +0530
@@ -605,6 +605,48 @@
 
 BLOCKCOPY_PP_W32_H4 32, 48
 
+%macro BLOCKCOPY_PP_W32_H8_avx 2
+INIT_YMM avx
+cglobal blockcopy_pp_%1x%2, 4, 7, 8
+    mov    r4d, %2/8
+    lea    r5,  [3 * r3]
+    lea    r6,  [3 * r1]
+
+.loop:
+    movu    m0, [r2]
+    movu    m1, [r2 + r3]
+    movu    m2, [r2 + 2 * r3]
+    movu    m3, [r2 + r5]
+    lea     r2, [r2 + 4 * r3]
+    movu    m4, [r2]
+    movu    m5, [r2 + r3]
+    movu    m6, [r2 + 2 * r3]
+    movu    m7, [r2 + r5]
+
+    movu    [r0], m0
+    movu    [r0 + r1], m1
+    movu    [r0 + 2 * r1], m2
+    movu    [r0 + r6], m3
+    lea     r0, [r0 + 4 * r1]
+    movu    [r0], m4
+    movu    [r0 + r1], m5
+    movu    [r0 + 2 * r1], m6
+    movu    [r0 + r6], m7
+
+    dec    r4d
+    lea    r0, [r0 + 4 * r1]
+    lea    r2, [r2 + 4 * r3]
+    jnz   .loop
+    RET
+%endmacro
+
+BLOCKCOPY_PP_W32_H8_avx 32, 8
+BLOCKCOPY_PP_W32_H8_avx 32, 16
+BLOCKCOPY_PP_W32_H8_avx 32, 24
+BLOCKCOPY_PP_W32_H8_avx 32, 32
+BLOCKCOPY_PP_W32_H8_avx 32, 48
+BLOCKCOPY_PP_W32_H8_avx 32, 64
+
 ;-----------------------------------------------------------------------------
 ; void blockcopy_pp_%1x%2(pixel *dest, intptr_t deststride, pixel *src, intptr_t srcstride)
 ;-----------------------------------------------------------------------------
diff -r 4680ab4f92b8 -r e441b1fe3f78 source/common/x86/blockcopy8.h
--- a/source/common/x86/blockcopy8.h	Thu Sep 18 18:16:25 2014 +0530
+++ b/source/common/x86/blockcopy8.h	Fri Sep 19 15:01:20 2014 +0530
@@ -183,6 +183,13 @@
 void x265_blockfill_s_16x16_sse2(int16_t *dst, intptr_t dstride, int16_t val);
 void x265_blockfill_s_32x32_sse2(int16_t *dst, intptr_t dstride, int16_t val);
 
+void x265_blockcopy_pp_32x8_avx(pixel * a, intptr_t stridea, pixel * b, intptr_t strideb);
+void x265_blockcopy_pp_32x16_avx(pixel * a, intptr_t stridea, pixel * b, intptr_t strideb);
+void x265_blockcopy_pp_32x24_avx(pixel * a, intptr_t stridea, pixel * b, intptr_t strideb);
+void x265_blockcopy_pp_32x48_avx(pixel * a, intptr_t stridea, pixel * b, intptr_t strideb);
+void x265_blockcopy_pp_32x32_avx(pixel * a, intptr_t stridea, pixel * b, intptr_t strideb);
+void x265_blockcopy_pp_32x64_avx(pixel * a, intptr_t stridea, pixel * b, intptr_t strideb);
+
 #undef BLOCKCOPY_COMMON
 #undef BLOCKCOPY_SS_PP
 #undef BLOCKCOPY_SP