Skip to content

F
ffmpeg.wasm-core
Commit 7538ad22 authored by James Almer's avatar James Almer Committed by Michael Niedermayer
Browse files
Options

x86/hevc_deblock: improve chroma functions register allocation 

Signed-off-by: 's avatarJames Almer <jamrial@gmail.com>
Signed-off-by: 's avatarMichael Niedermayer <michaelni@gmx.at>
parent 08c51e12
Show whitespace changes
Inline Side-by-side
Showing with 43 additions and 44 deletions
libavcodec/x86/hevc_deblock.asm
View file @ 7538ad22
...@@ -57,10 +57,10 @@ INIT_XMM sse2 ...@@ -57,10 +57,10 @@ INIT_XMM sse2
movd m4, %5 movd m4, %5
movd m6, %6 movd m6, %6
movd m5, %7 movd m5, %7
movd m7, %8 movd m3, %8
punpcklbw m4, m6 punpcklbw m4, m6
punpcklbw m5, m7 punpcklbw m5, m3
punpcklwd m4, m5 punpcklwd m4, m5
punpckhdq m2, m0, m4 punpckhdq m2, m0, m4
...@@ -120,10 +120,10 @@ INIT_XMM sse2 ...@@ -120,10 +120,10 @@ INIT_XMM sse2
movq m4, %5 movq m4, %5
movq m6, %6 movq m6, %6
movq m5, %7 movq m5, %7
movq m7, %8 movq m3, %8
punpcklwd m4, m6 punpcklwd m4, m6
punpcklwd m5, m7 punpcklwd m5, m3
punpckhdq m6, m4, m5 punpckhdq m6, m4, m5
punpckldq m4, m5 punpckldq m4, m5
...@@ -310,7 +310,7 @@ INIT_XMM sse2 ...@@ -310,7 +310,7 @@ INIT_XMM sse2
%endmacro %endmacro
ALIGN 16 ALIGN 16
; input in m0 ... m3 and tcs in r2. Output in m1 and m2 ; input in m0 ... m3 and tcs in tc (r2). Output in m1 and m2
%macro CHROMA_DEBLOCK_BODY 1 %macro CHROMA_DEBLOCK_BODY 1
psubw m4, m2, m1; q0 - p0 psubw m4, m2, m1; q0 - p0
psubw m5, m0, m3; p1 - q1 psubw m5, m0, m3; p1 - q1
...@@ -318,12 +318,11 @@ ALIGN 16 ...@@ -318,12 +318,11 @@ ALIGN 16
paddw m5, m4; paddw m5, m4;
;tc calculations ;tc calculations
movd m6, [r2]; tc0 movd m6, [tcq]; tc0
add r2, 4;
punpcklwd m6, m6 punpcklwd m6, m6
movd m7, [r2]; tc1 movd m4, [tcq+4]; tc1
punpcklwd m7, m7 punpcklwd m4, m4
shufps m6, m7, 0; tc0, tc1 shufps m6, m4, 0; tc0, tc1
pmullw m4, m6, [pw_m1]; -tc0, -tc1 pmullw m4, m6, [pw_m1]; -tc0, -tc1
;end tc calculations ;end tc calculations
...@@ -669,37 +668,37 @@ INIT_XMM sse2 ...@@ -669,37 +668,37 @@ INIT_XMM sse2
;----------------------------------------------------------------------------- ;-----------------------------------------------------------------------------
; void ff_hevc_v_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q) ; void ff_hevc_v_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q)
;----------------------------------------------------------------------------- ;-----------------------------------------------------------------------------
cglobal hevc_v_loop_filter_chroma_8, 3, 6, 8 cglobal hevc_v_loop_filter_chroma_8, 3, 5, 7, pix, stride, tc, pix0, r3stride
sub r0, 2 sub pixq, 2
lea r5, [3*r1] lea r3strideq, [3*strideq]
mov r4, r0 mov pix0q, pixq
add r0, r5 add pixq, r3strideq
TRANSPOSE4x8B_LOAD PASS8ROWS(r4, r0, r1, r5) TRANSPOSE4x8B_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
CHROMA_DEBLOCK_BODY 8 CHROMA_DEBLOCK_BODY 8
TRANSPOSE8x4B_STORE PASS8ROWS(r4, r0, r1, r5) TRANSPOSE8x4B_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq)
RET RET
cglobal hevc_v_loop_filter_chroma_10, 3, 6, 8 cglobal hevc_v_loop_filter_chroma_10, 3, 5, 7, pix, stride, tc, pix0, r3stride
sub r0, 4 sub pixq, 4
lea r5, [3*r1] lea r3strideq, [3*strideq]
mov r4, r0 mov pix0q, pixq
add r0, r5 add pixq, r3strideq
TRANSPOSE4x8W_LOAD PASS8ROWS(r4, r0, r1, r5) TRANSPOSE4x8W_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
CHROMA_DEBLOCK_BODY 10 CHROMA_DEBLOCK_BODY 10
TRANSPOSE8x4W_STORE PASS8ROWS(r4, r0, r1, r5) TRANSPOSE8x4W_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq)
RET RET
;----------------------------------------------------------------------------- ;-----------------------------------------------------------------------------
; void ff_hevc_h_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q ; void ff_hevc_h_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q
;----------------------------------------------------------------------------- ;-----------------------------------------------------------------------------
cglobal hevc_h_loop_filter_chroma_8, 3, 6, 8 cglobal hevc_h_loop_filter_chroma_8, 3, 4, 7, pix, stride, tc, pix0
mov r5, r0; pix mov pix0q, pixq
sub r5, r1 sub pix0q, strideq
sub r5, r1 sub pix0q, strideq
movq m0, [r5]; p1 movq m0, [pix0q]; p1
movq m1, [r5+r1]; p0 movq m1, [pix0q+strideq]; p0
movq m2, [r0]; q0 movq m2, [pixq]; q0
movq m3, [r0+r1]; q1 movq m3, [pixq+strideq]; q1
pxor m5, m5; zeros reg pxor m5, m5; zeros reg
punpcklbw m0, m5 punpcklbw m0, m5
punpcklbw m1, m5 punpcklbw m1, m5
...@@ -708,24 +707,24 @@ cglobal hevc_h_loop_filter_chroma_8, 3, 6, 8 ...@@ -708,24 +707,24 @@ cglobal hevc_h_loop_filter_chroma_8, 3, 6, 8
CHROMA_DEBLOCK_BODY 8 CHROMA_DEBLOCK_BODY 8
packuswb m1, m1 ; p0' packed in bytes on low quadword packuswb m1, m1 ; p0' packed in bytes on low quadword
packuswb m2, m2 ; q0' packed in bytes on low quadword packuswb m2, m2 ; q0' packed in bytes on low quadword
movq [r5+r1], m1 movq [pix0q+strideq], m1
movq [r0], m2 movq [pixq], m2
RET RET
cglobal hevc_h_loop_filter_chroma_10, 3, 6, 8 cglobal hevc_h_loop_filter_chroma_10, 3, 4, 7, pix, stride, tc, pix0
mov r5, r0; pix mov pix0q, pixq
sub r5, r1 sub pix0q, strideq
sub r5, r1 sub pix0q, strideq
movdqu m0, [r5]; p1 movu m0, [pix0q]; p1
movdqu m1, [r5+r1]; p0 movu m1, [pix0q+strideq]; p0
movdqu m2, [r0]; q0 movu m2, [pixq]; q0
movdqu m3, [r0+r1]; q1 movu m3, [pixq+strideq]; q1
CHROMA_DEBLOCK_BODY 10 CHROMA_DEBLOCK_BODY 10
pxor m5, m5; zeros reg pxor m5, m5; zeros reg
CLIPW m1, m5, [pw_pixel_max] CLIPW m1, m5, [pw_pixel_max]
CLIPW m2, m5, [pw_pixel_max] CLIPW m2, m5, [pw_pixel_max]
movdqu [r5+r1], m1 movu [pix0q+strideq], m1
movdqu [r0], m2 movu [pixq], m2
RET RET
%if ARCH_X86_64 %if ARCH_X86_64
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment