Move sse16_sse2() from inline asm to yasm. It is one of the functions causing

Win64/FATE issues.

Originally committed as revision 25136 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/x86/Makefile

@@ -24,6 +24,7 @@ MMX-OBJS-$(CONFIG_MP3FLOAT_DECODER)    += x86/mpegaudiodec_mmx.o
 MMX-OBJS-$(CONFIG_MP3ON4FLOAT_DECODER) += x86/mpegaudiodec_mmx.o
 MMX-OBJS-$(CONFIG_MP3ADUFLOAT_DECODER) += x86/mpegaudiodec_mmx.o
 MMX-OBJS-$(CONFIG_ENCODERS)            += x86/dsputilenc_mmx.o
+YASM-OBJS-$(CONFIG_ENCODERS)           += x86/dsputilenc_yasm.o
 MMX-OBJS-$(CONFIG_GPL)                 += x86/idct_mmx.o
 MMX-OBJS-$(CONFIG_LPC)                 += x86/lpc_mmx.o
 MMX-OBJS-$(CONFIG_DWT)                 += x86/snowdsp_mmx.o

libavcodec/x86/dsputilenc_mmx.c

@@ -323,67 +323,7 @@ static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int
     return tmp;
 }
 
-static int sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
-    int tmp;
-  __asm__ volatile (
-      "shr $1,%2\n"
-      "pxor %%xmm0,%%xmm0\n"    /* mm0 = 0 */
-      "pxor %%xmm7,%%xmm7\n"    /* mm7 holds the sum */
-      "1:\n"
-      "movdqu (%0),%%xmm1\n"    /* mm1 = pix1[0][0-15] */
-      "movdqu (%1),%%xmm2\n"    /* mm2 = pix2[0][0-15] */
-      "movdqu (%0,%4),%%xmm3\n" /* mm3 = pix1[1][0-15] */
-      "movdqu (%1,%4),%%xmm4\n" /* mm4 = pix2[1][0-15] */
-
-      /* todo: mm1-mm2, mm3-mm4 */
-      /* algo: subtract mm1 from mm2 with saturation and vice versa */
-      /*       OR the results to get absolute difference */
-      "movdqa %%xmm1,%%xmm5\n"
-      "movdqa %%xmm3,%%xmm6\n"
-      "psubusb %%xmm2,%%xmm1\n"
-      "psubusb %%xmm4,%%xmm3\n"
-      "psubusb %%xmm5,%%xmm2\n"
-      "psubusb %%xmm6,%%xmm4\n"
-
-      "por %%xmm1,%%xmm2\n"
-      "por %%xmm3,%%xmm4\n"
-
-      /* now convert to 16-bit vectors so we can square them */
-      "movdqa %%xmm2,%%xmm1\n"
-      "movdqa %%xmm4,%%xmm3\n"
-
-      "punpckhbw %%xmm0,%%xmm2\n"
-      "punpckhbw %%xmm0,%%xmm4\n"
-      "punpcklbw %%xmm0,%%xmm1\n"  /* mm1 now spread over (mm1,mm2) */
-      "punpcklbw %%xmm0,%%xmm3\n"  /* mm4 now spread over (mm3,mm4) */
-
-      "pmaddwd %%xmm2,%%xmm2\n"
-      "pmaddwd %%xmm4,%%xmm4\n"
-      "pmaddwd %%xmm1,%%xmm1\n"
-      "pmaddwd %%xmm3,%%xmm3\n"
-
-      "lea (%0,%4,2), %0\n"        /* pix1 += 2*line_size */
-      "lea (%1,%4,2), %1\n"        /* pix2 += 2*line_size */
-
-      "paddd %%xmm2,%%xmm1\n"
-      "paddd %%xmm4,%%xmm3\n"
-      "paddd %%xmm1,%%xmm7\n"
-      "paddd %%xmm3,%%xmm7\n"
-
-      "decl %2\n"
-      "jnz 1b\n"
-
-      "movdqa %%xmm7,%%xmm1\n"
-      "psrldq $8, %%xmm7\n"        /* shift hi qword to lo */
-      "paddd %%xmm1,%%xmm7\n"
-      "movdqa %%xmm7,%%xmm1\n"
-      "psrldq $4, %%xmm7\n"        /* shift hi dword to lo */
-      "paddd %%xmm1,%%xmm7\n"
-      "movd %%xmm7,%3\n"
-      : "+r" (pix1), "+r" (pix2), "+r"(h), "=r"(tmp)
-      : "r" ((x86_reg)line_size));
-    return tmp;
-}
+int ff_sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h);
 
 static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
     int tmp;
@@ -1376,7 +1316,7 @@ void dsputilenc_init_mmx(DSPContext* c, AVCodecContext *avctx)
         c->hadamard8_diff[1]= hadamard8_diff_mmx;
 
         c->pix_norm1 = pix_norm1_mmx;
-        c->sse[0] = (mm_flags & AV_CPU_FLAG_SSE2) ? sse16_sse2 : sse16_mmx;
+        c->sse[0] = (mm_flags & AV_CPU_FLAG_SSE2) ? ff_sse16_sse2 : sse16_mmx;
           c->sse[1] = sse8_mmx;
         c->vsad[4]= vsad_intra16_mmx;
 

libavcodec/x86/dsputilenc_yasm.asm

@@ -0,0 +1,87 @@
+;*****************************************************************************
+;* MMX optimized DSP utils
+;*****************************************************************************
+;* Copyright (c) 2000, 2001 Fabrice Bellard
+;* Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
+;*
+;* This file is part of FFmpeg.
+;*
+;* FFmpeg is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* FFmpeg is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with FFmpeg; if not, write to the Free Software
+;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;*****************************************************************************
+
+%include "x86inc.asm"
+%include "x86util.asm"
+
+SECTION .text
+
+INIT_XMM
+; sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h)
+cglobal sse16_sse2, 5, 5, 8
+    shr       r4, 1
+    pxor      m0, m0         ; mm0 = 0
+    pxor      m7, m7         ; mm7 holds the sum
+
+.next2lines ; FIXME why are these unaligned movs? pix1[] is aligned
+    movu      m1, [r1   ]    ; mm1 = pix1[0][0-15]
+    movu      m2, [r2   ]    ; mm2 = pix2[0][0-15]
+    movu      m3, [r1+r3]    ; mm3 = pix1[1][0-15]
+    movu      m4, [r2+r3]    ; mm4 = pix2[1][0-15]
+
+    ; todo: mm1-mm2, mm3-mm4
+    ; algo: subtract mm1 from mm2 with saturation and vice versa
+    ;       OR the result to get the absolute difference
+    mova      m5, m1
+    mova      m6, m3
+    psubusb   m1, m2
+    psubusb   m3, m4
+    psubusb   m2, m5
+    psubusb   m4, m6
+
+    por       m2, m1
+    por       m4, m3
+
+    ; now convert to 16-bit vectors so we can square them
+    mova      m1, m2
+    mova      m3, m4
+
+    punpckhbw m2, m0
+    punpckhbw m4, m0
+    punpcklbw m1, m0         ; mm1 not spread over (mm1,mm2)
+    punpcklbw m3, m0         ; mm4 not spread over (mm3,mm4)
+
+    pmaddwd   m2, m2
+    pmaddwd   m4, m4
+    pmaddwd   m1, m1
+    pmaddwd   m3, m3
+
+    lea       r1, [r1+r3*2]  ; pix1 += 2*line_size
+    lea       r2, [r2+r3*2]  ; pix2 += 2*line_size
+
+    paddd     m1, m2
+    paddd     m3, m4
+    paddd     m7, m1
+    paddd     m7, m3
+
+    dec       r4
+    jnz .next2lines
+
+    mova      m1, m7
+    psrldq    m7, 8          ; shift hi qword to lo
+    paddd     m7, m1
+    mova      m1, m7
+    psrldq    m7, 4          ; shift hi dword to lo
+    paddd     m7, m1
+    movd     eax, m7         ; return value
+    RET