[x265] [PATCH 3 of 6] asm: intra_pred_ang32_29 improved by ~58%, 3763.98c -> 1562.61c over SSE4

praveen at multicorewareinc.com praveen at multicorewareinc.com
Wed Mar 25 14:57:55 CET 2015


# HG changeset patch
# User Praveen Tiwari <praveen at multicorewareinc.com>
# Date 1427281060 -19800
#      Wed Mar 25 16:27:40 2015 +0530
# Node ID 40e9865d86a55fc195bf6efde2bb5f3422e36bdd
# Parent  6d41331ab23c01711354b0eadfc4d084b8dfc2df
asm: intra_pred_ang32_29 improved by ~58%, 3763.98c -> 1562.61c over SSE4

diff -r 6d41331ab23c -r 40e9865d86a5 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Wed Mar 25 16:21:04 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Wed Mar 25 16:27:40 2015 +0530
@@ -1603,6 +1603,7 @@
         p.cu[BLOCK_32x32].intra_pred[26] = x265_intra_pred_ang32_26_avx2;
         p.cu[BLOCK_32x32].intra_pred[27] = x265_intra_pred_ang32_27_avx2;
         p.cu[BLOCK_32x32].intra_pred[28] = x265_intra_pred_ang32_28_avx2;
+        p.cu[BLOCK_32x32].intra_pred[29] = x265_intra_pred_ang32_29_avx2;
 
         // copy_sp primitives
         p.cu[BLOCK_16x16].copy_sp = x265_blockcopy_sp_16x16_avx2;
diff -r 6d41331ab23c -r 40e9865d86a5 source/common/x86/intrapred.h
--- a/source/common/x86/intrapred.h	Wed Mar 25 16:21:04 2015 +0530
+++ b/source/common/x86/intrapred.h	Wed Mar 25 16:27:40 2015 +0530
@@ -208,6 +208,7 @@
 void x265_intra_pred_ang32_26_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang32_27_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang32_28_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
+void x265_intra_pred_ang32_29_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_all_angs_pred_4x4_sse4(pixel *dest, pixel *refPix, pixel *filtPix, int bLuma);
 void x265_all_angs_pred_8x8_sse4(pixel *dest, pixel *refPix, pixel *filtPix, int bLuma);
 void x265_all_angs_pred_16x16_sse4(pixel *dest, pixel *refPix, pixel *filtPix, int bLuma);
diff -r 6d41331ab23c -r 40e9865d86a5 source/common/x86/intrapred8.asm
--- a/source/common/x86/intrapred8.asm	Wed Mar 25 16:21:04 2015 +0530
+++ b/source/common/x86/intrapred8.asm	Wed Mar 25 16:27:40 2015 +0530
@@ -286,7 +286,27 @@
                     db 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17
                     db 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27
                     db 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0
-
+
+ALIGN 32
+c_ang32_mode_29:    db 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18
+                    db 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27
+                    db 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 19, 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 13
+                    db 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31
+                    db 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17
+                    db 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26
+                    db 29, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29, 3, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12
+                    db 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30
+                    db 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16
+                    db 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25
+                    db 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 21, 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 11
+                    db 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29
+                    db 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15
+                    db 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24
+                    db 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10
+                    db 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28
+                    db 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14
+                    db 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23
+                    db 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0
 
 ALIGN 32
 ;; (blkSize - 1 - x)
@@ -12084,5 +12104,277 @@
     vpermq            m6, m6, 11011000b
     movu              [r0 + r3], m6
     RET
+
+INIT_YMM avx2
+cglobal intra_pred_ang32_29, 3, 5, 11
+    mova              m0, [pw_1024]
+    mova              m1, [intra_pred_shuff_0_8]
+    lea               r3, [3 * r1]
+    lea               r4, [c_ang32_mode_29]
+
+    ;row [0, 1]
+    vbroadcasti128    m2, [r2 + 1]
+    pshufb            m2, m1
+    vbroadcasti128    m3, [r2 + 9]
+    pshufb            m3, m1
+    vbroadcasti128    m4, [r2 + 17]
+    pshufb            m4, m1
+    vbroadcasti128    m5, [r2 + 25]
+    pshufb            m5, m1
+
+    mova              m10, [r4 + 0 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0], m7
+    movu              [r0 + r1], m6
+
+    ;row [2]
+    mova              m10, [r4 + 1 * mmsize]
+    vperm2i128        m6, m2, m3, 00100000b
+    pmaddubsw         m6, m10
+    pmulhrsw          m6, m0
+    vperm2i128        m7, m4, m5, 00100000b
+    pmaddubsw         m7, m10
+    pmulhrsw          m7, m0
+    packuswb          m6, m7
+    vpermq            m6, m6, 11011000b
+    movu              [r0 + 2 * r1], m6
+
+    ;row [3, 4]
+    vbroadcasti128    m2, [r2 + 2]
+    pshufb            m2, m1
+    vbroadcasti128    m3, [r2 + 10]
+    pshufb            m3, m1
+    vbroadcasti128    m4, [r2 + 18]
+    pshufb            m4, m1
+    vbroadcasti128    m5, [r2 + 26]
+    pshufb            m5, m1
+
+    mova              m10, [r4 + 2 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0 + r3], m7
+    lea               r0, [r0 + 4 * r1]
+    movu              [r0], m6
+
+    ;row [5, 6]
+    mova              m10, [r4 + 3 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0 + r1], m7
+    movu              [r0 + 2 * r1], m6
+
+    ;row [7, 8]
+    vbroadcasti128    m2, [r2 + 3]
+    pshufb            m2, m1
+    vbroadcasti128    m3, [r2 + 11]
+    pshufb            m3, m1
+    vbroadcasti128    m4, [r2 + 19]
+    pshufb            m4, m1
+    vbroadcasti128    m5, [r2 + 27]
+    pshufb            m5, m1
+
+    add               r4, 4 * mmsize
+    mova              m10, [r4 + 0 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0 + r3], m7
+    lea               r0, [r0 + 4 * r1]
+    movu              [r0], m6
+
+    ;row [9]
+    mova              m10, [r4 + 1 * mmsize]
+    vperm2i128        m6, m2, m3, 00100000b
+    pmaddubsw         m6, m10
+    pmulhrsw          m6, m0
+    vperm2i128        m7, m4, m5, 00100000b
+    pmaddubsw         m7, m10
+    pmulhrsw          m7, m0
+    packuswb          m6, m7
+    vpermq            m6, m6, 11011000b
+    movu              [r0 + r1], m6
+
+    ;row [10, 11]
+    vbroadcasti128    m2, [r2 + 4]
+    pshufb            m2, m1
+    vbroadcasti128    m3, [r2 + 12]
+    pshufb            m3, m1
+    vbroadcasti128    m4, [r2 + 20]
+    pshufb            m4, m1
+    vbroadcasti128    m5, [r2 + 28]
+    pshufb            m5, m1
+
+    mova              m10, [r4 + 2 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0 + 2 * r1], m7
+    movu              [r0 + r3], m6
+
+    ;row [12, 13]
+    lea               r0, [r0 + 4 * r1]
+    mova              m10, [r4 + 3 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0], m7
+    movu              [r0 + r1], m6
+
+    ;row [14, 15]
+    vbroadcasti128    m2, [r2 + 5]
+    pshufb            m2, m1
+    vbroadcasti128    m3, [r2 + 13]
+    pshufb            m3, m1
+    vbroadcasti128    m4, [r2 + 21]
+    pshufb            m4, m1
+    vbroadcasti128    m5, [r2 + 29]
+    pshufb            m5, m1
+
+    add               r4, 4 * mmsize
+    mova              m10, [r4 + 0 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0 + 2 * r1], m7
+    movu              [r0 + r3], m6
+
+    ;row [16]
+    lea               r0, [r0 + 4 * r1]
+    mova              m10, [r4 + 1 * mmsize]
+    vperm2i128        m6, m2, m3, 00100000b
+    pmaddubsw         m6, m10
+    pmulhrsw          m6, m0
+    vperm2i128        m7, m4, m5, 00100000b
+    pmaddubsw         m7, m10
+    pmulhrsw          m7, m0
+    packuswb          m6, m7
+    vpermq            m6, m6, 11011000b
+    movu              [r0], m6
+
+    ;row [17, 18]
+    vbroadcasti128    m2, [r2 + 6]
+    pshufb            m2, m1
+    vbroadcasti128    m3, [r2 + 14]
+    pshufb            m3, m1
+    vbroadcasti128    m4, [r2 + 22]
+    pshufb            m4, m1
+    vbroadcasti128    m5, [r2 + 30]
+    pshufb            m5, m1
+
+    mova              m10, [r4 + 2 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0 + r1], m7
+    movu              [r0 + 2 * r1], m6
+
+    ;row [19, 20]
+    mova              m10, [r4 + 3 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0 + r3], m7
+    lea               r0, [r0 + 4 * r1]
+    movu              [r0], m6
+
+    ;row [21, 22]
+    vbroadcasti128    m2, [r2 + 7]
+    pshufb            m2, m1
+    vbroadcasti128    m3, [r2 + 15]
+    pshufb            m3, m1
+    vbroadcasti128    m4, [r2 + 23]
+    pshufb            m4, m1
+    vbroadcasti128    m5, [r2 + 31]
+    pshufb            m5, m1
+
+    add               r4, 4 * mmsize
+    mova              m10, [r4 + 0 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0 + r1], m7
+    movu              [r0 + 2 * r1], m6
+
+    ;row [23]
+    mova              m10, [r4 + 1 * mmsize]
+    vperm2i128        m6, m2, m3, 00100000b
+    pmaddubsw         m6, m10
+    pmulhrsw          m6, m0
+    vperm2i128        m7, m4, m5, 00100000b
+    pmaddubsw         m7, m10
+    pmulhrsw          m7, m0
+    packuswb          m6, m7
+    vpermq            m6, m6, 11011000b
+    movu              [r0 + r3], m6
+
+    ;row [24, 25]
+    vbroadcasti128    m2, [r2 + 8]
+    pshufb            m2, m1
+    vbroadcasti128    m3, [r2 + 16]
+    pshufb            m3, m1
+    vbroadcasti128    m4, [r2 + 24]
+    pshufb            m4, m1
+    vbroadcasti128    m5, [r2 + 32]
+    pshufb            m5, m1
+
+    lea               r0, [r0 + 4 * r1]
+    mova              m10, [r4 + 2 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0], m7
+    movu              [r0 + r1], m6
+
+    ;row [26, 27]
+    mova              m10, [r4 + 3 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0 + 2 * r1], m7
+    movu              [r0 + r3], m6
+
+    ;row [28, 29]
+    vbroadcasti128    m2, [r2 + 9]
+    pshufb            m2, m1
+    vbroadcasti128    m3, [r2 + 17]
+    pshufb            m3, m1
+    vbroadcasti128    m4, [r2 + 25]
+    pshufb            m4, m1
+    vbroadcasti128    m5, [r2 + 33]
+    pshufb            m5, m1
+
+    lea               r0, [r0 + 4 * r1]
+    add               r4, 4 * mmsize
+    mova              m10, [r4 + 0 * mmsize]
+
+    INTRA_PRED_ANG32_CAL_ROW
+    movu              [r0], m7
+    movu              [r0 + r1], m6
+
+    ;row [30]
+    mova              m10, [r4 + 1 * mmsize]
+    vperm2i128        m6, m2, m3, 00100000b
+    pmaddubsw         m6, m10
+    pmulhrsw          m6, m0
+    vperm2i128        m7, m4, m5, 00100000b
+    pmaddubsw         m7, m10
+    pmulhrsw          m7, m0
+    packuswb          m6, m7
+    vpermq            m6, m6, 11011000b
+    movu              [r0 + 2 * r1], m6
+
+    ;row [31]
+    vbroadcasti128    m2, [r2 + 10]
+    pshufb            m2, m1
+    vbroadcasti128    m3, [r2 + 18]
+    pshufb            m3, m1
+    vbroadcasti128    m4, [r2 + 26]
+    pshufb            m4, m1
+    vbroadcasti128    m5, [r2 + 34]
+    pshufb            m5, m1
+
+    mova              m10, [r4 + 2 * mmsize]
+    vperm2i128        m6, m2, m3, 00100000b
+    pmaddubsw         m6, m10
+    pmulhrsw          m6, m0
+    vperm2i128        m7, m4, m5, 00100000b
+    pmaddubsw         m7, m10
+    pmulhrsw          m7, m0
+    packuswb          m6, m7
+    vpermq            m6, m6, 11011000b
+    movu              [r0 + r3], m6
+    RET
 %endif
 


More information about the x265-devel mailing list