ref: 98fcccfe9751894ace9693a39ba0609fe5ea904d
parent: b2fa74ac18c4a333f7913a346588b87087989202
author: Jeff Muizelaar <jmuizelaar@mozilla.com>
date: Thu Jun 3 06:16:07 EDT 2010
Change the x86 idct functions to do reconstruction at the same time Change-Id: I896fe6f9664e6849c7cee2cc6bb4e045eb42540f
--- a/vp8/common/x86/idct_x86.h
+++ b/vp8/common/x86/idct_x86.h
@@ -22,6 +22,7 @@
#if HAVE_MMX
extern prototype_idct(vp8_short_idct4x4llm_1_mmx);
extern prototype_idct(vp8_short_idct4x4llm_mmx);
+extern prototype_idct_scalar_add(vp8_dc_only_idct_add_mmx);
extern prototype_second_order(vp8_short_inv_walsh4x4_mmx);
extern prototype_second_order(vp8_short_inv_walsh4x4_1_mmx);
@@ -32,6 +33,9 @@
#undef vp8_idct_idct16
#define vp8_idct_idct16 vp8_short_idct4x4llm_mmx
+
+#undef vp8_idct_idct1_scalar_add
+#define vp8_idct_idct1_scalar_add vp8_dc_only_idct_add_mmx
#undef vp8_idct_iwalsh16
#define vp8_idct_iwalsh16 vp8_short_inv_walsh4x4_mmx
--- a/vp8/common/x86/idctllm_mmx.asm
+++ b/vp8/common/x86/idctllm_mmx.asm
@@ -220,35 +220,61 @@
pop rbp
ret
-;void dc_only_idct_mmx(short input_dc, short *output, int pitch)
-global sym(vp8_dc_only_idct_mmx)
-sym(vp8_dc_only_idct_mmx):
+;void vp8_dc_only_idct_add_mmx(short input_dc, unsigned char *pred_ptr, unsigned char *dst_ptr, int pitch, int stride)
+global sym(vp8_dc_only_idct_add_mmx)
+sym(vp8_dc_only_idct_add_mmx):
push rbp
mov rbp, rsp
- SHADOW_ARGS_TO_STACK 3
+ SHADOW_ARGS_TO_STACK 5
GET_GOT rbx
+ push rsi
+ push rdi
; end prolog
- movd mm0, arg(0) ;input_dc
+ mov rsi, arg(1) ;s -- prediction
+ mov rdi, arg(2) ;d -- destination
+ movsxd rax, dword ptr arg(4) ;stride
+ movsxd rdx, dword ptr arg(3) ;pitch
+ pxor mm0, mm0
- paddw mm0, [fours GLOBAL]
- mov rdx, arg(1) ;output
+ movd mm5, arg(0) ;input_dc
- psraw mm0, 3
- movsxd rax, dword ptr arg(2) ;pitch
+ paddw mm5, [fours GLOBAL]
- punpcklwd mm0, mm0
- punpckldq mm0, mm0
+ psraw mm5, 3
- movq [rdx], mm0
- movq [rdx+rax], mm0
+ punpcklwd mm5, mm5
+ punpckldq mm5, mm5
- movq [rdx+rax*2], mm0
- add rdx, rax
+ movd mm1, [rsi]
+ punpcklbw mm1, mm0
+ paddsw mm1, mm5
+ packuswb mm1, mm0 ; pack and unpack to saturate
+ movd [rdi], mm1
- movq [rdx+rax*2], mm0
+ movd mm2, [rsi+rdx]
+ punpcklbw mm2, mm0
+ paddsw mm2, mm5
+ packuswb mm2, mm0 ; pack and unpack to saturate
+ movd [rdi+rax], mm2
+ movd mm3, [rsi+2*rdx]
+ punpcklbw mm3, mm0
+ paddsw mm3, mm5
+ packuswb mm3, mm0 ; pack and unpack to saturate
+ movd [rdi+2*rax], mm3
+
+ add rdi, rax
+ add rsi, rdx
+ movd mm4, [rsi+2*rdx]
+ punpcklbw mm4, mm0
+ paddsw mm4, mm5
+ packuswb mm4, mm0 ; pack and unpack to saturate
+ movd [rdi+2*rax], mm4
+
; begin epilog
+ pop rdi
+ pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
--- a/vp8/common/x86/x86_systemdependent.c
+++ b/vp8/common/x86/x86_systemdependent.c
@@ -42,6 +42,7 @@
{rtcd->idct.idct1 = vp8_short_idct4x4llm_1_mmx;
rtcd->idct.idct16 = vp8_short_idct4x4llm_mmx;
+ rtcd->idct.idct1_scalar_add = vp8_dc_only_idct_add_mmx;
rtcd->idct.iwalsh16 = vp8_short_inv_walsh4x4_mmx;
rtcd->idct.iwalsh1 = vp8_short_inv_walsh4x4_1_mmx;
--- a/vp8/decoder/x86/dequantize_mmx.asm
+++ b/vp8/decoder/x86/dequantize_mmx.asm
@@ -50,12 +50,12 @@
ret
-;void dequant_idct_mmx(short *input, short *dq, short *output, int pitch)
-global sym(vp8_dequant_idct_mmx)
-sym(vp8_dequant_idct_mmx):
+;void dequant_idct_add_mmx(short *input, short *dq, unsigned char *pred, unsigned char *dest, int pitch, int stride)
+global sym(vp8_dequant_idct_add_mmx)
+sym(vp8_dequant_idct_add_mmx):
push rbp
mov rbp, rsp
- SHADOW_ARGS_TO_STACK 4
+ SHADOW_ARGS_TO_STACK 6
GET_GOT rbx
push rsi
push rdi
@@ -77,7 +77,8 @@
movq mm3, [rax+24]
pmullw mm3, [rdx+24]
- mov rdx, arg(2) ;output
+ mov rdx, arg(3) ;dest
+ mov rsi, arg(2) ;pred
pxor mm7, mm7
@@ -88,7 +89,8 @@
movq [rax+24],mm7
- movsxd rax, dword ptr arg(3) ;pitch
+ movsxd rax, dword ptr arg(4) ;pitch
+ movsxd rdi, dword ptr arg(5) ;stride
psubw mm0, mm2 ; b1= 0-2
paddw mm2, mm2 ;
@@ -207,14 +209,35 @@
punpckldq mm2, mm4 ; 32 22 12 02
punpckhdq mm5, mm4 ; 33 23 13 03
- movq [rdx], mm0
+ pxor mm7, mm7
- movq [rdx+rax], mm1
- movq [rdx+rax*2], mm2
+ movd mm4, [rsi]
+ punpcklbw mm4, mm7
+ paddsw mm0, mm4
+ packuswb mm0, mm7
+ movd [rdx], mm0
- add rdx, rax
- movq [rdx+rax*2], mm5
+ movd mm4, [rsi+rax]
+ punpcklbw mm4, mm7
+ paddsw mm1, mm4
+ packuswb mm1, mm7
+ movd [rdx+rdi], mm1
+ movd mm4, [rsi+2*rax]
+ punpcklbw mm4, mm7
+ paddsw mm2, mm4
+ packuswb mm2, mm7
+ movd [rdx+rdi*2], mm2
+
+ add rdx, rdi
+ add rsi, rax
+
+ movd mm4, [rsi+2*rax]
+ punpcklbw mm4, mm7
+ paddsw mm5, mm4
+ packuswb mm5, mm7
+ movd [rdx+rdi*2], mm5
+
; begin epilog
pop rdi
pop rsi
@@ -224,12 +247,12 @@
ret
-;void dequant_dc_idct_mmx(short *input, short *dq, short *output, int pitch, int Dc)
-global sym(vp8_dequant_dc_idct_mmx)
-sym(vp8_dequant_dc_idct_mmx):
+;void dequant_dc_idct_add_mmx(short *input, short *dq, unsigned char *pred, unsigned char *dest, int pitch, int stride, int Dc)
+global sym(vp8_dequant_dc_idct_add_mmx)
+sym(vp8_dequant_dc_idct_add_mmx):
push rbp
mov rbp, rsp
- SHADOW_ARGS_TO_STACK 5
+ SHADOW_ARGS_TO_STACK 7
GET_GOT rbx
push rsi
push rdi
@@ -238,7 +261,7 @@
mov rax, arg(0) ;input
mov rdx, arg(1) ;dq
- movsxd rcx, dword ptr arg(4) ;Dc
+ movsxd rcx, dword ptr arg(6) ;Dc
movq mm0, [rax ]
pmullw mm0, [rdx]
@@ -252,7 +275,8 @@
movq mm3, [rax+24]
pmullw mm3, [rdx+24]
- mov rdx, arg(2) ;output
+ mov rdx, arg(3) ;dest
+ mov rsi, arg(2) ;pred
pxor mm7, mm7
@@ -262,8 +286,10 @@
movq [rax+16],mm7
movq [rax+24],mm7
+
pinsrw mm0, rcx, 0
- movsxd rax, dword ptr arg(3) ;pitch
+ movsxd rax, dword ptr arg(4) ;pitch
+ movsxd rdi, dword ptr arg(5) ;stride
psubw mm0, mm2 ; b1= 0-2
paddw mm2, mm2 ;
@@ -382,13 +408,34 @@
punpckldq mm2, mm4 ; 32 22 12 02
punpckhdq mm5, mm4 ; 33 23 13 03
- movq [rdx], mm0
+ pxor mm7, mm7
- movq [rdx+rax], mm1
- movq [rdx+rax*2], mm2
+ movd mm4, [rsi]
+ punpcklbw mm4, mm7
+ paddsw mm0, mm4
+ packuswb mm0, mm7
+ movd [rdx], mm0
- add rdx, rax
- movq [rdx+rax*2], mm5
+ movd mm4, [rsi+rax]
+ punpcklbw mm4, mm7
+ paddsw mm1, mm4
+ packuswb mm1, mm7
+ movd [rdx+rdi], mm1
+
+ movd mm4, [rsi+2*rax]
+ punpcklbw mm4, mm7
+ paddsw mm2, mm4
+ packuswb mm2, mm7
+ movd [rdx+rdi*2], mm2
+
+ add rdx, rdi
+ add rsi, rax
+
+ movd mm4, [rsi+2*rax]
+ punpcklbw mm4, mm7
+ paddsw mm5, mm4
+ packuswb mm5, mm7
+ movd [rdx+rdi*2], mm5
; begin epilog
pop rdi
--- a/vp8/decoder/x86/dequantize_x86.h
+++ b/vp8/decoder/x86/dequantize_x86.h
@@ -21,11 +21,19 @@
*/
#if HAVE_MMX
extern prototype_dequant_block(vp8_dequantize_b_mmx);
+extern prototype_dequant_idct_add(vp8_dequant_idct_add_mmx);
+extern prototype_dequant_idct_dc_add(vp8_dequant_dc_idct_add_mmx);
#if !CONFIG_RUNTIME_CPU_DETECT
#undef vp8_dequant_block
#define vp8_dequant_block vp8_dequantize_b_mmx
+
+#undef vp8_dequant_idct_add
+#define vp8_dequant_idct_add vp8_dequant_idct_add_mmx
+
+#undef vp8_dequant_idct_dc
+#define vp8_dequant_idct_add_dc vp8_dequant_dc_idct_add_mmx
#endif
#endif
--- a/vp8/decoder/x86/x86_dsystemdependent.c
+++ b/vp8/decoder/x86/x86_dsystemdependent.c
@@ -43,6 +43,8 @@
if (flags & HAS_MMX)
{pbi->dequant.block = vp8_dequantize_b_mmx;
+ pbi->dequant.idct_add = vp8_dequant_idct_add_mmx;
+ pbi->dequant.idct_dc_add = vp8_dequant_dc_idct_add_mmx;
}
#endif