[x265] [PATCH] asm: luma_vpp[32x24] in avx2: improve 5562c->3899c

[Divya Manivannan]

# HG changeset patch
# User Divya Manivannan <divya at multicorewareinc.com>
# Date 1416897152 -19800
#      Tue Nov 25 12:02:32 2014 +0530
# Node ID 083389c66af862aa102db76ff6f697e206df3205
# Parent  d7fbfb0a0d4c959904ec149fded360d78261f60d
asm: luma_vpp[32x24] in avx2: improve 5562c->3899c

diff -r d7fbfb0a0d4c -r 083389c66af8 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Tue Nov 25 11:45:44 2014 +0530
+++ b/source/common/x86/asm-primitives.cpp	Tue Nov 25 12:02:32 2014 +0530
@@ -1806,6 +1806,7 @@
         p.luma_vpp[LUMA_16x64] = x265_interp_8tap_vert_pp_16x64_avx2;
         p.luma_vpp[LUMA_32x8] = x265_interp_8tap_vert_pp_32x8_avx2;
         p.luma_vpp[LUMA_32x16] = x265_interp_8tap_vert_pp_32x16_avx2;
+        p.luma_vpp[LUMA_32x24] = x265_interp_8tap_vert_pp_32x24_avx2;
         p.luma_vpp[LUMA_32x32] = x265_interp_8tap_vert_pp_32x32_avx2;
         p.luma_vpp[LUMA_32x64] = x265_interp_8tap_vert_pp_32x64_avx2;
 #endif
diff -r d7fbfb0a0d4c -r 083389c66af8 source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm	Tue Nov 25 11:45:44 2014 +0530
+++ b/source/common/x86/ipfilter8.asm	Tue Nov 25 12:02:32 2014 +0530
@@ -5574,6 +5574,44 @@
     RET
 %endif
 
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_pp_32x24, 4, 10, 15
+    mov             r4d, r4m
+    shl             r4d, 7
+
+%ifdef PIC
+    lea             r5, [tab_LumaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_LumaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r4
+    lea             r6, [r3 * 3]
+    mova            m14, [pw_512]
+    mov             r9d, 2
+.loopW:
+    PROCESS_LUMA_AVX2_W16_16R
+    add             r2, 16
+    add             r0, 16
+    dec             r9d
+    jnz             .loopW
+    lea             r9, [r1 * 4]
+    sub             r7, r9
+    lea             r0, [r7 - 16]
+    lea             r2, [r8 + r3 * 4 - 16]
+    mov             r9d, 2
+.loop:
+    PROCESS_LUMA_AVX2_W16_8R
+    add             r2, 16
+    add             r0, 16
+    dec             r9d
+    jnz             .loop
+    RET
+%endif
+
 ;-------------------------------------------------------------------------------------------------------------
 ; void interp_8tap_vert_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
 ;-------------------------------------------------------------------------------------------------------------