[x265] [PATCH] asm: avx2 8bpp code for chroma_p2s[16xN] for i420, i422, improved over 25% than SSE

[rajesh at multicorewareinc.com]

# HG changeset patch
# User Rajesh Paulraj<rajesh at multicorewareinc.com>
# Date 1442307006 -19800
#      Tue Sep 15 14:20:06 2015 +0530
# Node ID 4fe52a4c976193e8ddbeb16b5d3edeeac4fab6c8
# Parent  31d45b06c3ee6d747c8d4f50f171aa6479b1c60f
asm: avx2 8bpp code for chroma_p2s[16xN] for i420, i422, improved over 25% than SSE

diff -r 31d45b06c3ee -r 4fe52a4c9761 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Tue Sep 15 13:56:14 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Tue Sep 15 14:20:06 2015 +0530
@@ -3305,11 +3305,21 @@
         p.pu[LUMA_48x64].convert_p2s = PFX(filterPixelToShort_48x64_avx2);
         p.pu[LUMA_24x32].convert_p2s = PFX(filterPixelToShort_24x32_avx2);
 
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_16x4].p2s = PFX(filterPixelToShort_16x4_avx2);
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_16x8].p2s = PFX(filterPixelToShort_16x8_avx2);
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_16x12].p2s = PFX(filterPixelToShort_16x12_avx2);
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_16x16].p2s = PFX(filterPixelToShort_16x16_avx2);
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_16x32].p2s = PFX(filterPixelToShort_16x32_avx2);
         p.chroma[X265_CSP_I420].pu[CHROMA_420_24x32].p2s = PFX(filterPixelToShort_24x32_avx2);
         p.chroma[X265_CSP_I420].pu[CHROMA_420_32x8].p2s = PFX(filterPixelToShort_32x8_avx2);
         p.chroma[X265_CSP_I420].pu[CHROMA_420_32x16].p2s = PFX(filterPixelToShort_32x16_avx2);
         p.chroma[X265_CSP_I420].pu[CHROMA_420_32x24].p2s = PFX(filterPixelToShort_32x24_avx2);
         p.chroma[X265_CSP_I420].pu[CHROMA_420_32x32].p2s = PFX(filterPixelToShort_32x32_avx2);
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x8].p2s = PFX(filterPixelToShort_16x8_avx2);
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x16].p2s = PFX(filterPixelToShort_16x16_avx2);
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x24].p2s = PFX(filterPixelToShort_16x24_avx2);
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x32].p2s = PFX(filterPixelToShort_16x32_avx2);
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x64].p2s = PFX(filterPixelToShort_16x64_avx2);
         p.chroma[X265_CSP_I422].pu[CHROMA_422_24x64].p2s = PFX(filterPixelToShort_24x64_avx2);
         p.chroma[X265_CSP_I422].pu[CHROMA_422_32x16].p2s = PFX(filterPixelToShort_32x16_avx2);
         p.chroma[X265_CSP_I422].pu[CHROMA_422_32x32].p2s = PFX(filterPixelToShort_32x32_avx2);
diff -r 31d45b06c3ee -r 4fe52a4c9761 source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm	Tue Sep 15 13:56:14 2015 +0530
+++ b/source/common/x86/ipfilter8.asm	Tue Sep 15 14:20:06 2015 +0530
@@ -12816,6 +12816,70 @@
 ;-----------------------------------------------------------------------------
 ; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
+INIT_YMM avx2
+cglobal filterPixelToShort_16x24, 3, 7, 2
+    mov             r3d, r3m
+    add             r3d, r3d
+    lea             r4, [r1 * 3]
+    lea             r5, [r3 * 3]
+    mov             r6d, 3
+
+    ; load constant
+    vbroadcasti128  m1, [pw_2000]
+.loop:
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
+
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
+
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
+
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
+
+    pmovzxbw        m0, [r0]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2], m0
+
+    pmovzxbw        m0, [r0 + r1]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3], m0
+
+    pmovzxbw        m0, [r0 + r1 * 2]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r3 * 2], m0
+
+    pmovzxbw        m0, [r0 + r4]
+    psllw           m0, 6
+    psubw           m0, m1
+    movu            [r2 + r5], m0
+
+    lea             r0, [r0 + r1 * 4]
+    lea             r2, [r2 + r3 * 4]
+
+    dec             r6d
+    jnz             .loop
+    RET
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
 %macro P2S_H_16xN_avx2 1
 INIT_YMM avx2
 cglobal filterPixelToShort_16x%1, 3, 7, 2