[x265] [PATCH 1 of 2] asm: saoCuOrgE3 asm code
nabajit at multicorewareinc.com
nabajit at multicorewareinc.com
Wed Jan 7 10:00:16 CET 2015
# HG changeset patch
# User Nabajit Deka
# Date 1420620491 -19800
# Wed Jan 07 14:18:11 2015 +0530
# Node ID 9ec89f245be8ca4468362cb095172dbc92bd5140
# Parent 6cc757f662ed982a2f64122eba8e557d8ef0abba
asm: saoCuOrgE3 asm code
diff -r 6cc757f662ed -r 9ec89f245be8 source/common/loopfilter.cpp
--- a/source/common/loopfilter.cpp Wed Jan 07 13:44:23 2015 +0530
+++ b/source/common/loopfilter.cpp Wed Jan 07 14:18:11 2015 +0530
@@ -87,6 +87,22 @@
}
}
+void processSaoCUE3(pixel *rec, int8_t *upBuff1, int8_t *m_offsetEo, intptr_t stride, int startX, int endX)
+ {
+ int8_t signDown;
+ int8_t edgeType;
+
+ for (int x = startX + 1; x < endX; x++)
+ {
+ signDown = signOf(rec[x] - rec[x + stride]);
+ edgeType = signDown + upBuff1[x] + 2;
+ upBuff1[x - 1] = -signDown;
+
+ short v = rec[x] + m_offsetEo[edgeType];
+ rec[x] = (pixel)(v < 0 ? 0 : (v > (PIXEL_MAX)) ? (PIXEL_MAX) : v);
+ }
+ }
+
void processSaoCUB0(pixel* rec, const int8_t* offset, int ctuWidth, int ctuHeight, intptr_t stride)
{
#define SAO_BO_BITS 5
@@ -113,6 +129,7 @@
p.saoCuOrgE0 = processSaoCUE0;
p.saoCuOrgE1 = processSaoCUE1;
p.saoCuOrgE2 = processSaoCUE2;
+ p.saoCuOrgE3 = processSaoCUE3;
p.saoCuOrgB0 = processSaoCUB0;
p.sign = calSign;
}
diff -r 6cc757f662ed -r 9ec89f245be8 source/common/primitives.h
--- a/source/common/primitives.h Wed Jan 07 13:44:23 2015 +0530
+++ b/source/common/primitives.h Wed Jan 07 14:18:11 2015 +0530
@@ -193,6 +193,7 @@
typedef void (*saoCuOrgE0_t)(pixel* rec, int8_t* offsetEo, int width, int8_t signLeft);
typedef void (*saoCuOrgE1_t)(pixel* rec, int8_t* upBuff1, int8_t* offsetEo, intptr_t stride, int width);
typedef void (*saoCuOrgE2_t)(pixel* rec, int8_t* pBufft, int8_t* pBuff1, int8_t* offsetEo, int lcuWidth, intptr_t stride);
+typedef void (*saoCuOrgE3_t)(pixel* rec, int8_t* upBuff1, int8_t* m_offsetEo, intptr_t stride, int startX, int endX);
typedef void (*saoCuOrgB0_t)(pixel* rec, const int8_t* offsetBo, int ctuWidth, int ctuHeight, intptr_t stride);
typedef void (*sign_t)(int8_t *dst, const pixel *src1, const pixel *src2, const int endX);
typedef void (*planecopy_cp_t) (const uint8_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int width, int height, int shift);
@@ -248,6 +249,7 @@
saoCuOrgE0_t saoCuOrgE0;
saoCuOrgE1_t saoCuOrgE1;
saoCuOrgE2_t saoCuOrgE2;
+ saoCuOrgE3_t saoCuOrgE3;
saoCuOrgB0_t saoCuOrgB0;
downscale_t frameInitLowres;
diff -r 6cc757f662ed -r 9ec89f245be8 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Wed Jan 07 13:44:23 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp Wed Jan 07 14:18:11 2015 +0530
@@ -1652,6 +1652,7 @@
p.saoCuOrgE0 = x265_saoCuOrgE0_sse4;
p.saoCuOrgE1 = x265_saoCuOrgE1_sse4;
p.saoCuOrgE2 = x265_saoCuOrgE2_sse4;
+ p.saoCuOrgE3 = x265_saoCuOrgE3_sse4;
p.saoCuOrgB0 = x265_saoCuOrgB0_sse4;
LUMA_ADDAVG(_sse4);
diff -r 6cc757f662ed -r 9ec89f245be8 source/common/x86/loopfilter.asm
--- a/source/common/x86/loopfilter.asm Wed Jan 07 13:44:23 2015 +0530
+++ b/source/common/x86/loopfilter.asm Wed Jan 07 14:18:11 2015 +0530
@@ -188,6 +188,88 @@
jnz .loop
RET
+;=======================================================================================================
+;void saoCuOrgE3(pixel *rec, int8_t *upBuff1, int8_t *m_offsetEo, intptr_t stride, int startX, int endX)
+;=======================================================================================================
+INIT_XMM sse4
+cglobal saoCuOrgE3, 3, 7, 8
+ mov r3d, r3m
+ mov r4d, r4m
+ mov r5d, r5m
+
+ mov r6d, r5d
+ sub r6d, r4d
+
+ inc r4d
+ add r0, r4
+ add r1, r4
+ movh m7, [r0 + r6 - 1]
+ mov r6, [r1 + r6 - 2]
+ pxor m0, m0 ; m0 = 0
+ movu m6, [pb_2] ; m6 = [2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2]
+
+.loop:
+ movu m1, [r0] ; m1 = pRec[x]
+ movu m2, [r0 + r3] ; m2 = pRec[x + iStride]
+
+ psubusb m3, m2, m1
+ psubusb m4, m1, m2
+ pcmpeqb m3, m0
+ pcmpeqb m4, m0
+ pcmpeqb m2, m1
+
+ pabsb m3, m3
+ por m4, m3
+ pandn m2, m4 ; m2 = iSignDown
+
+ movu m3, [r1] ; m3 = m_iUpBuff1
+
+ paddb m3, m2
+ paddb m3, m6 ; m3 = uiEdgeType
+
+ movu m4, [r2] ; m4 = m_iOffsetEo
+ pshufb m5, m4, m3
+
+ psubb m3, m0, m2
+ movu [r1 - 1], m3
+
+ pmovzxbw m2, m1
+ punpckhbw m1, m0
+ pmovsxbw m3, m5
+ punpckhbw m5, m5
+ psraw m5, 8
+
+ paddw m2, m3
+ paddw m1, m5
+ packuswb m2, m1
+ movu [r0], m2
+
+ sub r5d, 16
+ jle .end
+
+ lea r0, [r0 + 16]
+ lea r1, [r1 + 16]
+
+ jnz .loop
+
+.end:
+ js .skip
+ sub r0, r4
+ sub r1, r4
+ movh [r0 + 16], m7
+ mov [r1 + 15], r6
+ jmp .quit
+
+.skip:
+ sub r0, r4
+ sub r1, r4
+ movh [r0 + 15], m7
+ mov [r1 + 14], r6
+
+.quit:
+
+ RET
+
;=====================================================================================
; void saoCuOrgB0(pixel* rec, const pixel* offset, int lcuWidth, int lcuHeight, int stride)
;=====================================================================================
diff -r 6cc757f662ed -r 9ec89f245be8 source/common/x86/loopfilter.h
--- a/source/common/x86/loopfilter.h Wed Jan 07 13:44:23 2015 +0530
+++ b/source/common/x86/loopfilter.h Wed Jan 07 14:18:11 2015 +0530
@@ -28,6 +28,7 @@
void x265_saoCuOrgE0_sse4(pixel * rec, int8_t * offsetEo, int endX, int8_t signLeft);
void x265_saoCuOrgE1_sse4(pixel* rec, int8_t* upBuff1, int8_t* offsetEo, intptr_t stride, int width);
void x265_saoCuOrgE2_sse4(pixel* rec, int8_t* pBufft, int8_t* pBuff1, int8_t* offsetEo, int lcuWidth, intptr_t stride);
+void x265_saoCuOrgE3_sse4(pixel *rec, int8_t *upBuff1, int8_t *m_offsetEo, intptr_t stride, int startX, int endX);
void x265_saoCuOrgB0_sse4(pixel* rec, const int8_t* offsetBo, int ctuWidth, int ctuHeight, intptr_t stride);
void x265_calSign_sse4(int8_t *dst, const pixel *src1, const pixel *src2, const int endX);
diff -r 6cc757f662ed -r 9ec89f245be8 source/test/pixelharness.cpp
--- a/source/test/pixelharness.cpp Wed Jan 07 13:44:23 2015 +0530
+++ b/source/test/pixelharness.cpp Wed Jan 07 14:18:11 2015 +0530
@@ -978,6 +978,35 @@
return true;
}
+bool PixelHarness::check_saoCuOrgE3_t(saoCuOrgE3_t ref, saoCuOrgE3_t opt)
+{
+ ALIGN_VAR_16(pixel, ref_dest[64 * 64]);
+ ALIGN_VAR_16(pixel, opt_dest[64 * 64]);
+
+ memset(ref_dest, 0xCD, sizeof(ref_dest));
+ memset(opt_dest, 0xCD, sizeof(opt_dest));
+
+ int j = 0;
+
+ for (int i = 0; i < ITERS; i++)
+ {
+ int stride = 16 * (rand() % 4 + 1);
+ int start = rand() % 2;
+ int end = (16 * (rand() % 4 + 1)) - rand() % 2;
+
+ ref(ref_dest, psbuf2 + j, psbuf1 + j, stride, start, end);
+ checked(opt, opt_dest, psbuf5 + j, psbuf1 + j, stride, start, end);
+
+ if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel)) || memcmp(psbuf2, psbuf5, BUFFSIZE))
+ return false;
+
+ reportfail();
+ j += INCR;
+ }
+
+ return true;
+}
+
bool PixelHarness::check_planecopy_sp(planecopy_sp_t ref, planecopy_sp_t opt)
{
ALIGN_VAR_16(pixel, ref_dest[64 * 64]);
@@ -1542,6 +1571,15 @@
}
}
+ if (opt.saoCuOrgE3)
+ {
+ if (!check_saoCuOrgE3_t(ref.saoCuOrgE3, opt.saoCuOrgE3))
+ {
+ printf("SAO_EO_3 failed\n");
+ return false;
+ }
+ }
+
if (opt.saoCuOrgB0)
{
if (!check_saoCuOrgB0_t(ref.saoCuOrgB0, opt.saoCuOrgB0))
@@ -1892,6 +1930,12 @@
REPORT_SPEEDUP(opt.saoCuOrgE2, ref.saoCuOrgE2, pbuf1, psbuf1, psbuf2, psbuf3, 64, 64);
}
+ if (opt.saoCuOrgE3)
+ {
+ HEADER0("SAO_EO_3");
+ REPORT_SPEEDUP(opt.saoCuOrgE3, ref.saoCuOrgE3, pbuf1, psbuf2, psbuf1, 64, 0, 64);
+ }
+
if (opt.saoCuOrgB0)
{
HEADER0("SAO_BO_0");
diff -r 6cc757f662ed -r 9ec89f245be8 source/test/pixelharness.h
--- a/source/test/pixelharness.h Wed Jan 07 13:44:23 2015 +0530
+++ b/source/test/pixelharness.h Wed Jan 07 14:18:11 2015 +0530
@@ -96,6 +96,7 @@
bool check_saoCuOrgE0_t(saoCuOrgE0_t ref, saoCuOrgE0_t opt);
bool check_saoCuOrgE1_t(saoCuOrgE1_t ref, saoCuOrgE1_t opt);
bool check_saoCuOrgE2_t(saoCuOrgE2_t ref, saoCuOrgE2_t opt);
+ bool check_saoCuOrgE3_t(saoCuOrgE3_t ref, saoCuOrgE3_t opt);
bool check_saoCuOrgB0_t(saoCuOrgB0_t ref, saoCuOrgB0_t opt);
bool check_planecopy_sp(planecopy_sp_t ref, planecopy_sp_t opt);
bool check_planecopy_cp(planecopy_cp_t ref, planecopy_cp_t opt);
More information about the x265-devel
mailing list