summaryrefslogtreecommitdiff
path: root/libavcodec/x86/ac3dsp.asm
diff options
context:
space:
mode:
authorJustin Ruggles <justin.ruggles@gmail.com>2011-03-11 16:45:01 -0500
committerJustin Ruggles <justin.ruggles@gmail.com>2011-03-14 08:45:31 -0400
commitf1efbca5e914b28cd37cb1ea82d371acf3b96f61 (patch)
treea1464d5951e558f17e941c9f13d3548b733e6c03 /libavcodec/x86/ac3dsp.asm
parent323e6fead07c75f418e4b60704a4f437bb3483b2 (diff)
ac3enc: add SIMD-optimized shifting functions for use with the fixed-point AC3 encoder.
Diffstat (limited to 'libavcodec/x86/ac3dsp.asm')
-rw-r--r--libavcodec/x86/ac3dsp.asm45
1 files changed, 45 insertions, 0 deletions
diff --git a/libavcodec/x86/ac3dsp.asm b/libavcodec/x86/ac3dsp.asm
index b1eeca9cd2..e281791b1e 100644
--- a/libavcodec/x86/ac3dsp.asm
+++ b/libavcodec/x86/ac3dsp.asm
@@ -133,3 +133,48 @@ INIT_XMM
AC3_MAX_MSB_ABS_INT16 sse2, min_max
%define ABS2 ABS2_SSSE3
AC3_MAX_MSB_ABS_INT16 ssse3, or_abs
+
+;-----------------------------------------------------------------------------
+; macro used for ff_ac3_lshift_int16() and ff_ac3_rshift_int32()
+;-----------------------------------------------------------------------------
+
+%macro AC3_SHIFT 4 ; l/r, 16/32, shift instruction, instruction set
+cglobal ac3_%1shift_int%2_%4, 3,3,5, src, len, shift
+ movd m0, shiftd
+.loop:
+ mova m1, [srcq ]
+ mova m2, [srcq+mmsize ]
+ mova m3, [srcq+mmsize*2]
+ mova m4, [srcq+mmsize*3]
+ %3 m1, m0
+ %3 m2, m0
+ %3 m3, m0
+ %3 m4, m0
+ mova [srcq ], m1
+ mova [srcq+mmsize ], m2
+ mova [srcq+mmsize*2], m3
+ mova [srcq+mmsize*3], m4
+ add srcq, mmsize*4
+ sub lend, mmsize*32/%2
+ ja .loop
+.end:
+ REP_RET
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void ff_ac3_lshift_int16(int16_t *src, unsigned int len, unsigned int shift)
+;-----------------------------------------------------------------------------
+
+INIT_MMX
+AC3_SHIFT l, 16, psllw, mmx
+INIT_XMM
+AC3_SHIFT l, 16, psllw, sse2
+
+;-----------------------------------------------------------------------------
+; void ff_ac3_rshift_int32(int32_t *src, unsigned int len, unsigned int shift)
+;-----------------------------------------------------------------------------
+
+INIT_MMX
+AC3_SHIFT r, 32, psrad, mmx
+INIT_XMM
+AC3_SHIFT r, 32, psrad, sse2