[FFmpeg-devel] [PATCH 1/3] x86: simple_idct10_template: use const

Christophe Gisquet christophe.gisquet at gmail.com
Tue Oct 13 21:21:39 CEST 2015


This avoid going through constants.c while still sharing them
with proresdsp.asm
---
 libavcodec/x86/constants.c                | 28 ----------------------------
 libavcodec/x86/constants.h                | 16 ----------------
 libavcodec/x86/proresdsp.asm              | 13 +++++++++++++
 libavcodec/x86/simple_idct10.asm          | 26 ++++++++++++++++++++++++++
 libavcodec/x86/simple_idct10_template.asm | 13 -------------
 5 files changed, 39 insertions(+), 57 deletions(-)

diff --git a/libavcodec/x86/constants.c b/libavcodec/x86/constants.c
index 7e3d490..0098e20 100644
--- a/libavcodec/x86/constants.c
+++ b/libavcodec/x86/constants.c
@@ -89,31 +89,3 @@ DECLARE_ALIGNED(32, const ymm_reg,  ff_pd_8192) = { 0x0000200000002000ULL, 0x000
                                                     0x0000200000002000ULL, 0x0000200000002000ULL };
 DECLARE_ALIGNED(32, const ymm_reg,  ff_pd_65535)= { 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL,
                                                     0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL };
-
-/* simple idct 10 */
-#define W1sh2 22725 // W1 = 90901 = 22725<<2 + 1
-#define W2sh2 21407 // W2 = 85627 = 21407<<2 - 1
-#define W3sh2 19265 // W3 = 77062 = 19265<<2 + 2
-#define W4sh2 16384 // W4 = 65535 = 16384<<2 - 1
-#define W5sh2 12873 // W5 = 51491 = 12873<<2 - 1
-#define W6sh2  8867 // W6 = 35468 =  8867<<2
-#define W7sh2  4520 // W7 = 18081 =  4520<<2 + 1
-
-#define TIMES4(a, b)                                                        \
-     (((b)&0xFFFFLL)<<48)+(((a)&0xFFFFLL)<<32)+(((b)&0xFFFFLL)<<16)+((a)&0xFFFFLL), \
-     (((b)&0xFFFFLL)<<48)+(((a)&0xFFFFLL)<<32)+(((b)&0xFFFFLL)<<16)+((a)&0xFFFFLL)
-
-#if ARCH_X86_64
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w4_plus_w2) = { TIMES4(W4sh2,  W2sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w4_min_w2)  = { TIMES4(W4sh2, -W2sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w4_plus_w6) = { TIMES4(W4sh2, +W6sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w4_min_w6)  = { TIMES4(W4sh2, -W6sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w1_plus_w3) = { TIMES4(W1sh2, +W3sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w3_min_w1)  = { TIMES4(W3sh2, -W1sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w7_plus_w3) = { TIMES4(W7sh2, +W3sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w3_min_w7)  = { TIMES4(W3sh2, -W7sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w1_plus_w5) = { TIMES4(W1sh2, +W5sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w5_min_w1)  = { TIMES4(W5sh2, -W1sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w5_plus_w7) = { TIMES4(W5sh2, +W7sh2) };
-DECLARE_ALIGNED(16, const xmm_reg,  ff_w7_min_w5)  = { TIMES4(W7sh2, -W5sh2) };
-#endif
diff --git a/libavcodec/x86/constants.h b/libavcodec/x86/constants.h
index 496933f..f989755 100644
--- a/libavcodec/x86/constants.h
+++ b/libavcodec/x86/constants.h
@@ -68,20 +68,4 @@ extern const ymm_reg  ff_pd_32;
 extern const ymm_reg  ff_pd_8192;
 extern const ymm_reg  ff_pd_65535;
 
-# if ARCH_X86_64
-/* simple_idct10, used by prores and dnxhd */
-extern const xmm_reg  ff_w4_plus_w2;
-extern const xmm_reg  ff_w4_min_w2;
-extern const xmm_reg  ff_w4_plus_w6;
-extern const xmm_reg  ff_w4_min_w6;
-extern const xmm_reg  ff_w1_plus_w3;
-extern const xmm_reg  ff_w3_min_w1;
-extern const xmm_reg  ff_w7_plus_w3;
-extern const xmm_reg  ff_w3_min_w7;
-extern const xmm_reg  ff_w1_plus_w5;
-extern const xmm_reg  ff_w5_min_w1;
-extern const xmm_reg  ff_w5_plus_w7;
-extern const xmm_reg  ff_w7_min_w5;
-# endif
-
 #endif /* AVCODEC_X86_CONSTANTS_H */
diff --git a/libavcodec/x86/proresdsp.asm b/libavcodec/x86/proresdsp.asm
index 4633844..edfd2f7 100644
--- a/libavcodec/x86/proresdsp.asm
+++ b/libavcodec/x86/proresdsp.asm
@@ -32,6 +32,19 @@ pw_88:      times 8 dw 0x2008
 cextern pw_1
 cextern pw_4
 cextern pw_1019
+; Below are defined in simple_idct10.asm built from selecting idctdsp
+cextern w4_plus_w2
+cextern w4_min_w2
+cextern w4_plus_w6
+cextern w4_min_w6
+cextern w1_plus_w3
+cextern w3_min_w1
+cextern w7_plus_w3
+cextern w3_min_w7
+cextern w1_plus_w5
+cextern w5_min_w1
+cextern w5_plus_w7
+cextern w7_min_w5
 
 %include "libavcodec/x86/simple_idct10_template.asm"
 
diff --git a/libavcodec/x86/simple_idct10.asm b/libavcodec/x86/simple_idct10.asm
index ec388f9..cd83d61 100644
--- a/libavcodec/x86/simple_idct10.asm
+++ b/libavcodec/x86/simple_idct10.asm
@@ -37,6 +37,32 @@ pd_round_12: times 4 dd 1<<(12-1)
 pd_round_15: times 4 dd 1<<(15-1)
 pd_round_19: times 4 dd 1<<(19-1)
 
+%macro CONST_DEC  3
+const %1
+times 4 dw %2, %3
+%endmacro
+
+%define W1sh2 22725 ; W1 = 90901 = 22725<<2 + 1
+%define W2sh2 21407 ; W2 = 85627 = 21407<<2 - 1
+%define W3sh2 19265 ; W3 = 77062 = 19265<<2 + 2
+%define W4sh2 16384 ; W4 = 65535 = 16384<<2 - 1
+%define W5sh2 12873 ; W5 = 51491 = 12873<<2 - 1
+%define W6sh2  8867 ; W6 = 35468 =  8867<<2
+%define W7sh2  4520 ; W7 = 18081 =  4520<<2 + 1
+
+CONST_DEC  w4_plus_w2,   W4sh2, +W2sh2
+CONST_DEC  w4_min_w2,    W4sh2, -W2sh2
+CONST_DEC  w4_plus_w6,   W4sh2, +W6sh2
+CONST_DEC  w4_min_w6,    W4sh2, -W6sh2
+CONST_DEC  w1_plus_w3,   W1sh2, +W3sh2
+CONST_DEC  w3_min_w1,    W3sh2, -W1sh2
+CONST_DEC  w7_plus_w3,   W7sh2, +W3sh2
+CONST_DEC  w3_min_w7,    W3sh2, -W7sh2
+CONST_DEC  w1_plus_w5,   W1sh2, +W5sh2
+CONST_DEC  w5_min_w1,    W5sh2, -W1sh2
+CONST_DEC  w5_plus_w7,   W5sh2, +W7sh2
+CONST_DEC  w7_min_w5,    W7sh2, -W5sh2
+
 %include "libavcodec/x86/simple_idct10_template.asm"
 
 section .text align=16
diff --git a/libavcodec/x86/simple_idct10_template.asm b/libavcodec/x86/simple_idct10_template.asm
index 7300c8a..e5deb0f 100644
--- a/libavcodec/x86/simple_idct10_template.asm
+++ b/libavcodec/x86/simple_idct10_template.asm
@@ -26,19 +26,6 @@
 
 %if ARCH_X86_64
 
-cextern w4_plus_w2
-cextern w4_min_w2
-cextern w4_plus_w6
-cextern w4_min_w6
-cextern w1_plus_w3
-cextern w3_min_w1
-cextern w7_plus_w3
-cextern w3_min_w7
-cextern w1_plus_w5
-cextern w5_min_w1
-cextern w5_plus_w7
-cextern w7_min_w5
-
 ; interleave data while maintaining source
 ; %1=type, %2=dstlo, %3=dsthi, %4=src, %5=interleave
 %macro SBUTTERFLY3 5
-- 
2.6.0



More information about the ffmpeg-devel mailing list