From bbe4a6db44f0b55b424a5cc9d3e89cd88e250450 Mon Sep 17 00:00:00 2001 From: Henrik Gramner Date: Wed, 11 Sep 2013 17:49:26 +0200 Subject: x86inc: Utilize the shadow space on 64-bit Windows Store XMM6 and XMM7 in the shadow space in functions that clobbers them. This way we don't have to adjust the stack pointer as often, reducing the number of instructions as well as code size. Signed-off-by: Derek Buitenhuis --- libavcodec/x86/fft.asm | 4 ++-- libavcodec/x86/h264_deblock.asm | 19 +++++++------------ 2 files changed, 9 insertions(+), 14 deletions(-) (limited to 'libavcodec') diff --git a/libavcodec/x86/fft.asm b/libavcodec/x86/fft.asm index c87752bae8..e4744a3b60 100644 --- a/libavcodec/x86/fft.asm +++ b/libavcodec/x86/fft.asm @@ -667,13 +667,13 @@ cglobal imdct_calc, 3,5,3 push r1 push r0 %else - sub rsp, 8 + sub rsp, 8+32*WIN64 ; allocate win64 shadow space %endif call r4 %if ARCH_X86_32 add esp, 12 %else - add rsp, 8 + add rsp, 8+32*WIN64 %endif POP r1 POP r3 diff --git a/libavcodec/x86/h264_deblock.asm b/libavcodec/x86/h264_deblock.asm index fc6c983052..6e29ce7373 100644 --- a/libavcodec/x86/h264_deblock.asm +++ b/libavcodec/x86/h264_deblock.asm @@ -331,16 +331,14 @@ cglobal deblock_v_luma_8, 5,5,10 ; void deblock_h_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 ) ;----------------------------------------------------------------------------- INIT_MMX cpuname -cglobal deblock_h_luma_8, 5,9 +cglobal deblock_h_luma_8, 5,9,0,0x60+16*WIN64 movsxd r7, r1d lea r8, [r7+r7*2] lea r6, [r0-4] lea r5, [r0-4+r8] %if WIN64 - sub rsp, 0x98 - %define pix_tmp rsp+0x30 + %define pix_tmp rsp+0x30 ; shadow space + r4 %else - sub rsp, 0x68 %define pix_tmp rsp %endif @@ -379,11 +377,6 @@ cglobal deblock_h_luma_8, 5,9 movq m3, [pix_tmp+0x40] TRANSPOSE8x4B_STORE PASS8ROWS(r6, r5, r7, r8) -%if WIN64 - add rsp, 0x98 -%else - add rsp, 0x68 -%endif RET %endmacro @@ -704,13 +697,16 @@ INIT_MMX cpuname ;----------------------------------------------------------------------------- ; void deblock_h_luma_intra( uint8_t *pix, int stride, int alpha, int beta ) ;----------------------------------------------------------------------------- -cglobal deblock_h_luma_intra_8, 4,9 +cglobal deblock_h_luma_intra_8, 4,9,0,0x80 movsxd r7, r1d lea r8, [r7*3] lea r6, [r0-4] lea r5, [r0-4+r8] - sub rsp, 0x88 +%if WIN64 + %define pix_tmp rsp+0x20 ; shadow space +%else %define pix_tmp rsp +%endif ; transpose 8x16 -> tmp space TRANSPOSE8x8_MEM PASS8ROWS(r6, r5, r7, r8), PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30) @@ -730,7 +726,6 @@ cglobal deblock_h_luma_intra_8, 4,9 sub r5, r7 shr r7, 3 TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30), PASS8ROWS(r6, r5, r7, r8) - add rsp, 0x88 RET %else cglobal deblock_h_luma_intra_8, 2,4,8,0x80 -- cgit v1.2.3