Commit fc380de0 authored by Derk-Jan Hartman's avatar Derk-Jan Hartman

* Part of the Intel patches are now fixed upstream

parent 04870f7c
......@@ -154,172 +154,6 @@ diff -u -r1.22 dsputil_mmx_rnd.h
"1: \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t"
"movq 1(%1, %%"REG_a"), %%mm2 \n\t"
Index: libavcodec/i386/fdct_mmx.c
===================================================================
RCS file: /cvsroot/ffmpeg/ffmpeg/libavcodec/i386/fdct_mmx.c,v
retrieving revision 1.19
diff -u -r1.19 fdct_mmx.c
--- libavcodec/i386/fdct_mmx.c 22 Dec 2005 01:10:09 -0000 1.19
+++ libavcodec/i386/fdct_mmx.c 26 Jan 2006 22:02:55 -0000
@@ -350,61 +350,67 @@
static always_inline void fdct_row_sse2(const int16_t *in, int16_t *out)
{
+#define FDCT_ROW_SSE2_H1(i,t) \
+ "movq " #i "(%0), %%xmm2 \n\t" \
+ "movq " #i "+8(%0), %%xmm0 \n\t" \
+ "movdqa " #t "+32(%1), %%xmm3 \n\t" \
+ "movdqa " #t "+48(%1), %%xmm7 \n\t" \
+ "movdqa " #t "(%1), %%xmm4 \n\t" \
+ "movdqa " #t "+16(%1), %%xmm5 \n\t"
+
+#define FDCT_ROW_SSE2_H2(i,t) \
+ "movq " #i "(%0), %%xmm2 \n\t" \
+ "movq " #i "+8(%0), %%xmm0 \n\t" \
+ "movdqa " #t "+32(%1), %%xmm3 \n\t" \
+ "movdqa " #t "+48(%1), %%xmm7 \n\t"
+
+#define FDCT_ROW_SSE2(i) \
+ "movq %%xmm2, %%xmm1 \n\t" \
+ "pshuflw $27, %%xmm0, %%xmm0 \n\t" \
+ "paddsw %%xmm0, %%xmm1 \n\t" \
+ "psubsw %%xmm0, %%xmm2 \n\t" \
+ "punpckldq %%xmm2, %%xmm1 \n\t" \
+ "pshufd $78, %%xmm1, %%xmm2 \n\t" \
+ "pmaddwd %%xmm2, %%xmm3 \n\t" \
+ "pmaddwd %%xmm1, %%xmm7 \n\t" \
+ "pmaddwd %%xmm5, %%xmm2 \n\t" \
+ "pmaddwd %%xmm4, %%xmm1 \n\t" \
+ "paddd %%xmm7, %%xmm3 \n\t" \
+ "paddd %%xmm2, %%xmm1 \n\t" \
+ "paddd %%xmm6, %%xmm3 \n\t" \
+ "paddd %%xmm6, %%xmm1 \n\t" \
+ "psrad %3, %%xmm3 \n\t" \
+ "psrad %3, %%xmm1 \n\t" \
+ "packssdw %%xmm3, %%xmm1 \n\t" \
+ "movdqa %%xmm1, " #i "(%4) \n\t"
+
asm volatile(
- ".macro FDCT_ROW_SSE2_H1 i t \n\t"
- "movq \\i(%0), %%xmm2 \n\t"
- "movq \\i+8(%0), %%xmm0 \n\t"
- "movdqa \\t+32(%1), %%xmm3 \n\t"
- "movdqa \\t+48(%1), %%xmm7 \n\t"
- "movdqa \\t(%1), %%xmm4 \n\t"
- "movdqa \\t+16(%1), %%xmm5 \n\t"
- ".endm \n\t"
- ".macro FDCT_ROW_SSE2_H2 i t \n\t"
- "movq \\i(%0), %%xmm2 \n\t"
- "movq \\i+8(%0), %%xmm0 \n\t"
- "movdqa \\t+32(%1), %%xmm3 \n\t"
- "movdqa \\t+48(%1), %%xmm7 \n\t"
- ".endm \n\t"
- ".macro FDCT_ROW_SSE2 i \n\t"
- "movq %%xmm2, %%xmm1 \n\t"
- "pshuflw $27, %%xmm0, %%xmm0 \n\t"
- "paddsw %%xmm0, %%xmm1 \n\t"
- "psubsw %%xmm0, %%xmm2 \n\t"
- "punpckldq %%xmm2, %%xmm1 \n\t"
- "pshufd $78, %%xmm1, %%xmm2 \n\t"
- "pmaddwd %%xmm2, %%xmm3 \n\t"
- "pmaddwd %%xmm1, %%xmm7 \n\t"
- "pmaddwd %%xmm5, %%xmm2 \n\t"
- "pmaddwd %%xmm4, %%xmm1 \n\t"
- "paddd %%xmm7, %%xmm3 \n\t"
- "paddd %%xmm2, %%xmm1 \n\t"
- "paddd %%xmm6, %%xmm3 \n\t"
- "paddd %%xmm6, %%xmm1 \n\t"
- "psrad %3, %%xmm3 \n\t"
- "psrad %3, %%xmm1 \n\t"
- "packssdw %%xmm3, %%xmm1 \n\t"
- "movdqa %%xmm1, \\i(%4) \n\t"
- ".endm \n\t"
+// ".macro FDCT_ROW_SSE2_H1 i t \n\t"
+// ".endm \n\t"
+// ".macro FDCT_ROW_SSE2_H2 i t \n\t"
+// ".endm \n\t"
+// ".macro FDCT_ROW_SSE2 i \n\t"
+// ".endm \n\t"
"movdqa (%2), %%xmm6 \n\t"
- "FDCT_ROW_SSE2_H1 0 0 \n\t"
- "FDCT_ROW_SSE2 0 \n\t"
- "FDCT_ROW_SSE2_H2 64 0 \n\t"
- "FDCT_ROW_SSE2 64 \n\t"
-
- "FDCT_ROW_SSE2_H1 16 64 \n\t"
- "FDCT_ROW_SSE2 16 \n\t"
- "FDCT_ROW_SSE2_H2 112 64 \n\t"
- "FDCT_ROW_SSE2 112 \n\t"
-
- "FDCT_ROW_SSE2_H1 32 128 \n\t"
- "FDCT_ROW_SSE2 32 \n\t"
- "FDCT_ROW_SSE2_H2 96 128 \n\t"
- "FDCT_ROW_SSE2 96 \n\t"
-
- "FDCT_ROW_SSE2_H1 48 192 \n\t"
- "FDCT_ROW_SSE2 48 \n\t"
- "FDCT_ROW_SSE2_H2 80 192 \n\t"
- "FDCT_ROW_SSE2 80 \n\t"
+ FDCT_ROW_SSE2_H1(0,0)
+ FDCT_ROW_SSE2(0)
+ FDCT_ROW_SSE2_H2(64,0)
+ FDCT_ROW_SSE2(64)
+
+ FDCT_ROW_SSE2_H1(16,64)
+ FDCT_ROW_SSE2(16)
+ FDCT_ROW_SSE2_H2(112,64)
+ FDCT_ROW_SSE2(112)
+
+ FDCT_ROW_SSE2_H1(32,128)
+ FDCT_ROW_SSE2(32)
+ FDCT_ROW_SSE2_H2(96,128)
+ FDCT_ROW_SSE2(96)
+
+ FDCT_ROW_SSE2_H1(48,192)
+ FDCT_ROW_SSE2(48)
+ FDCT_ROW_SSE2_H2(80,192)
+ FDCT_ROW_SSE2(80)
:
: "r" (in), "r" (tab_frw_01234567_sse2.tab_frw_01234567_sse2), "r" (fdct_r_row_sse2.fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out)
);
Index: libavcodec/i386/idct_mmx_xvid.c
===================================================================
RCS file: /cvsroot/ffmpeg/ffmpeg/libavcodec/i386/idct_mmx_xvid.c,v
retrieving revision 1.4
diff -u -r1.4 idct_mmx_xvid.c
--- libavcodec/i386/idct_mmx_xvid.c 12 Jan 2006 22:43:18 -0000 1.4
+++ libavcodec/i386/idct_mmx_xvid.c 26 Jan 2006 22:02:55 -0000
@@ -295,17 +295,17 @@
"movq 8+" #A1 ",%%mm1 \n\t"/* 1 ; x7 x6 x5 x4*/\
"movq %%mm0,%%mm2 \n\t"/* 2 ; x3 x2 x1 x0*/\
"movq " #A3 ",%%mm3 \n\t"/* 3 ; w05 w04 w01 w00*/\
- "pshufw $0b10001000,%%mm0,%%mm0 \n\t"/* x2 x0 x2 x0*/\
+ "pshufw $0x88,%%mm0,%%mm0 \n\t"/* x2 x0 x2 x0*/\
"movq 8+" #A3 ",%%mm4 \n\t"/* 4 ; w07 w06 w03 w02*/\
"movq %%mm1,%%mm5 \n\t"/* 5 ; x7 x6 x5 x4*/\
"pmaddwd %%mm0,%%mm3 \n\t"/* x2*w05+x0*w04 x2*w01+x0*w00*/\
"movq 32+" #A3 ",%%mm6 \n\t"/* 6 ; w21 w20 w17 w16*/\
- "pshufw $0b10001000,%%mm1,%%mm1 \n\t"/* x6 x4 x6 x4*/\
+ "pshufw $0x88,%%mm1,%%mm1 \n\t"/* x6 x4 x6 x4*/\
"pmaddwd %%mm1,%%mm4 \n\t"/* x6*w07+x4*w06 x6*w03+x4*w02*/\
"movq 40+" #A3 ",%%mm7 \n\t"/* 7 ; w23 w22 w19 w18*/\
- "pshufw $0b11011101,%%mm2,%%mm2 \n\t"/* x3 x1 x3 x1*/\
+ "pshufw $0xdd,%%mm2,%%mm2 \n\t"/* x3 x1 x3 x1*/\
"pmaddwd %%mm2,%%mm6 \n\t"/* x3*w21+x1*w20 x3*w17+x1*w16*/\
- "pshufw $0b11011101,%%mm5,%%mm5 \n\t"/* x7 x5 x7 x5*/\
+ "pshufw $0xdd,%%mm5,%%mm5 \n\t"/* x7 x5 x7 x5*/\
"pmaddwd %%mm5,%%mm7 \n\t"/* x7*w23+x5*w22 x7*w19+x5*w18*/\
"paddd " #A4 ",%%mm3 \n\t"/* +%4*/\
"pmaddwd 16+" #A3 ",%%mm0 \n\t"/* x2*w13+x0*w12 x2*w09+x0*w08*/\
@@ -330,7 +330,7 @@
"packssdw %%mm0,%%mm3 \n\t"/* 0 ; y3 y2 y1 y0*/\
"packssdw %%mm4,%%mm7 \n\t"/* 4 ; y6 y7 y4 y5*/\
"movq %%mm3, " #A2 " \n\t"/* 3 ; save y3 y2 y1 y0*/\
- "pshufw $0b10110001,%%mm7,%%mm7 \n\t"/* y7 y6 y5 y4*/\
+ "pshufw $0xb1,%%mm7,%%mm7 \n\t"/* y7 y6 y5 y4*/\
"movq %%mm7,8 +" #A2 "\n\t"/* 7 ; save y7 y6 y5 y4*/\
Index: libavcodec/i386/motion_est_mmx.c
===================================================================
RCS file: /cvsroot/ffmpeg/ffmpeg/libavcodec/i386/motion_est_mmx.c,v
......
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