#define CC_TYPE "$cc_type"
#define CC_VERSION $cc_version
#define restrict $_restrict
-#define ASMALIGN(ZEROBITS) ".p2align " #ZEROBITS "\\n\\t"
#define EXTERN_PREFIX "${extern_prefix}"
#define EXTERN_ASM ${extern_prefix}
#define SLIBSUF "$SLIBSUF"
DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 };
DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };
-#define JUMPALIGN() __asm__ volatile (ASMALIGN(3)::)
+#define JUMPALIGN() __asm__ volatile (".p2align 3"::)
#define MOVQ_ZERO(regd) __asm__ volatile ("pxor %%" #regd ", %%" #regd ::)
#define MOVQ_BFE(regd) \
{
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movd (%1), %%mm0 \n\t"
"movd (%1, %3), %%mm1 \n\t"
{
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm1 \n\t"
{
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq 8(%1), %%mm4 \n\t"
"lea (%3, %3), %%"REG_a" \n\t"
"movq (%1), %%mm0 \n\t"
PAVGB" 1(%1), %%mm0 \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1, %%"REG_a"), %%mm2 \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movd %4, %%mm5 \n\t"
"punpcklwd %%mm5, %%mm5 \n\t"
"punpcklwd %%mm5, %%mm5 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%1, %0), %%mm0 \n\t"
"movq 8(%1, %0), %%mm1 \n\t"
"movd %3, %%mm5 \n\t"
"punpcklwd %%mm5, %%mm5 \n\t"
"punpcklwd %%mm5, %%mm5 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%1, %0), %%mm0 \n\t"
"movq 8(%1, %0), %%mm1 \n\t"
MOVQ_BFE(mm6);
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq 1(%1), %%mm1 \n\t"
"movq %%mm4, (%3) \n\t"
"add %5, %3 \n\t"
"decl %0 \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%2), %%mm1 \n\t"
MOVQ_BFE(mm6);
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq 1(%1), %%mm1 \n\t"
"movq %%mm5, 8(%3) \n\t"
"add %5, %3 \n\t"
"decl %0 \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%2), %%mm1 \n\t"
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"movq (%1), %%mm0 \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq (%1, %%"REG_a"),%%mm2 \n\t"
"paddusw %%mm1, %%mm5 \n\t"
"xor %%"REG_a", %%"REG_a" \n\t"
"add %3, %1 \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t"
"movq 1(%1, %%"REG_a"), %%mm2 \n\t"
__asm__ volatile(
"lea (%3, %3), %%"REG_a" \n\t"
"movq (%1), %%mm0 \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1, %3), %%mm1 \n\t"
"movq (%1, %%"REG_a"), %%mm2 \n\t"
"paddusw %%mm1, %%mm5 \n\t"
"xor %%"REG_a", %%"REG_a" \n\t"
"add %3, %1 \n\t"
- ASMALIGN(3)
+ ".p2align 3 \n\t"
"1: \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t"
"movq 1(%1, %%"REG_a"), %%mm2 \n\t"
__asm__ volatile(
"mov $-128, %%"REG_a" \n\t"
"pxor %%mm7, %%mm7 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%0), %%mm0 \n\t"
"movq (%0, %2), %%mm2 \n\t"
__asm__ volatile(
"pxor %%mm7, %%mm7 \n\t"
"mov $-128, %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%0), %%mm0 \n\t"
"movq (%1), %%mm2 \n\t"
TEST_TWO_ROWS("5*16(%0)", "6*16(%0)", "%%eax", "%%edx", CLEAR_ODD(ROW5), CLEAR_EVEN(ROW6))
TEST_ONE_ROW("7*16(%0)", "%%esi", CLEAR_ODD(ROW7))
iLLM_HEAD
- ASMALIGN(4)
+ ".p2align 4 \n\t"
JNZ("%%ecx", "2f")
JNZ("%%eax", "3f")
JNZ("%%edx", "4f")
{
x86_reg len= -(stride*h);
__asm__ volatile(
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t"
"movq (%2, %%"REG_a"), %%mm2 \n\t"
static inline void sad8_1_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
{
__asm__ volatile(
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm1 \n\t"
int ret;
__asm__ volatile(
"pxor %%xmm2, %%xmm2 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movdqu (%1), %%xmm0 \n\t"
"movdqu (%1, %4), %%xmm1 \n\t"
static inline void sad8_x2a_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
{
__asm__ volatile(
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%1), %%mm0 \n\t"
"movq (%1, %3), %%mm1 \n\t"
__asm__ volatile(
"movq (%1), %%mm0 \n\t"
"add %3, %1 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%1), %%mm1 \n\t"
"movq (%1, %3), %%mm2 \n\t"
"movq (%1), %%mm0 \n\t"
"pavgb 1(%1), %%mm0 \n\t"
"add %3, %1 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%1), %%mm1 \n\t"
"movq (%1,%3), %%mm2 \n\t"
{
x86_reg len= -(stride*h);
__asm__ volatile(
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t"
"movq (%2, %%"REG_a"), %%mm1 \n\t"
"punpckhbw %%mm7, %%mm3 \n\t"
"paddw %%mm2, %%mm0 \n\t"
"paddw %%mm3, %%mm1 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%2, %%"REG_a"), %%mm2 \n\t"
"movq 1(%2, %%"REG_a"), %%mm4 \n\t"
"packssdw %%mm5, %%mm5 \n\t"
"psubw %%mm5, %%mm7 \n\t"
"pxor %%mm4, %%mm4 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%0, %3), %%mm0 \n\t"
"movq 8(%0, %3), %%mm1 \n\t"
"packssdw %%mm5, %%mm5 \n\t"
"psubw %%mm5, %%mm7 \n\t"
"pxor %%mm4, %%mm4 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%0, %3), %%mm0 \n\t"
"movq 8(%0, %3), %%mm1 \n\t"
"packssdw %%mm6, %%mm6 \n\t"
"packssdw %%mm6, %%mm6 \n\t"
"mov %3, %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%0, %%"REG_a"), %%mm0 \n\t"
"movq 8(%0, %%"REG_a"), %%mm1 \n\t"
"packssdw %%mm6, %%mm6 \n\t"
"packssdw %%mm6, %%mm6 \n\t"
"mov %3, %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%0, %%"REG_a"), %%mm0 \n\t"
"movq 8(%0, %%"REG_a"), %%mm1 \n\t"
"packssdw %%mm6, %%mm6 \n\t"
"packssdw %%mm6, %%mm6 \n\t"
"mov %3, %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%0, %%"REG_a"), %%mm0 \n\t"
"movq 8(%0, %%"REG_a"), %%mm1 \n\t"
"packssdw %%mm6, %%mm6 \n\t"
"packssdw %%mm6, %%mm6 \n\t"
"mov %3, %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movq (%0, %%"REG_a"), %%mm0 \n\t"
"movq 8(%0, %%"REG_a"), %%mm1 \n\t"
"pxor "MM"6, "MM"6 \n\t"
"psubw (%3), "MM"6 \n\t" // -bias[0]
"mov $-128, %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
MOVQ" (%1, %%"REG_a"), "MM"0 \n\t" // block[i]
SAVE_SIGN(MM"1", MM"0") // ABS(block[i])
"pxor "MM"7, "MM"7 \n\t" // 0
"pxor "MM"4, "MM"4 \n\t" // 0
"mov $-128, %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
MOVQ" (%1, %%"REG_a"), "MM"0 \n\t" // block[i]
SAVE_SIGN(MM"1", MM"0") // ABS(block[i])
IDCT( 24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
"jmp 9f \n\t"
- "#" ASMALIGN(4) \
+ "# .p2align 4 \n\t"\
"4: \n\t"
Z_COND_IDCT( 64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11, 6f)
Z_COND_IDCT( 96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 5f)
IDCT( 24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
"jmp 9f \n\t"
- "#" ASMALIGN(4) \
+ "# .p2align 4 \n\t"\
"6: \n\t"
Z_COND_IDCT( 96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 7f)
IDCT( 24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
"jmp 9f \n\t"
- "#" ASMALIGN(4) \
+ "# .p2align 4 \n\t"\
"2: \n\t"
Z_COND_IDCT( 96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 3f)
IDCT( 24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
"jmp 9f \n\t"
- "#" ASMALIGN(4) \
+ "# .p2align 4 \n\t"\
"3: \n\t"
#undef IDCT
#define IDCT(src0, src4, src1, src5, dst, shift) \
IDCT( 24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
"jmp 9f \n\t"
- "#" ASMALIGN(4) \
+ "# .p2align 4 \n\t"\
"5: \n\t"
#undef IDCT
#define IDCT(src0, src4, src1, src5, dst, shift) \
"jmp 9f \n\t"
- "#" ASMALIGN(4) \
+ "# .p2align 4 \n\t"\
"1: \n\t"
#undef IDCT
#define IDCT(src0, src4, src1, src5, dst, shift) \
"jmp 9f \n\t"
- "#" ASMALIGN(4)
+ "# .p2align 4 \n\t"
"7: \n\t"
#undef IDCT
#define IDCT(src0, src4, src1, src5, dst, shift) \
LOAD_ROUNDER_MMX("%5") \
"movq "MANGLE(ff_pw_53)", %%mm5\n\t" \
"movq "MANGLE(ff_pw_18)", %%mm6\n\t" \
- ASMALIGN(3) \
+ ".p2align 3 \n\t" \
"1: \n\t" \
MSPEL_FILTER13_CORE(DO_UNPACK, "movd 1", A1, A2, A3, A4) \
NORMALIZE_MMX("%6") \
LOAD_ROUNDER_MMX("%4") \
"movq "MANGLE(ff_pw_18)", %%mm6 \n\t" \
"movq "MANGLE(ff_pw_53)", %%mm5 \n\t" \
- ASMALIGN(3) \
+ ".p2align 3 \n\t" \
"1: \n\t" \
MSPEL_FILTER13_CORE(DONT_UNPACK, "movq 2", A1, A2, A3, A4) \
NORMALIZE_MMX("$7") \
LOAD_ROUNDER_MMX("%6") \
"movq "MANGLE(ff_pw_53)", %%mm5 \n\t" \
"movq "MANGLE(ff_pw_18)", %%mm6 \n\t" \
- ASMALIGN(3) \
+ ".p2align 3 \n\t" \
"1: \n\t" \
MSPEL_FILTER13_CORE(DO_UNPACK, "movd 1", A1, A2, A3, A4) \
NORMALIZE_MMX("$6") \
"movq %4, %%mm6 \n\t"
"movq %5, %%mm7 \n\t"
"jmp 2f \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 32(%1) \n\t"
"movd (%1), %%mm0 \n\t"
"movq %4, %%mm6 \n\t"
"movq %5, %%mm7 \n\t"
"jmp 2f \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 32(%1) \n\t"
"movd (%1), %%mm0 \n\t"
"pxor %4, %%mm7 \n\t"
"movq %%mm7, %%mm6 \n\t"
"pxor %5, %%mm7 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 32(%1, %0) \n\t"
"movq (%1, %0), %%mm0 \n\t"
"movq "MANGLE(mask24r)", %%mm5 \n\t"
"movq "MANGLE(mask24g)", %%mm6 \n\t"
"movq "MANGLE(mask24b)", %%mm7 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 32(%1, %%"REG_a") \n\t"
"movq (%1, %%"REG_a"), %%mm0 \n\t" // BGR BGR BG
//FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway)
__asm__ volatile(
"xor %%"REG_a", %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 32(%1, %%"REG_a", 2) \n\t"
PREFETCH" 32(%2, %%"REG_a") \n\t"
//FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway)
__asm__ volatile(
"xor %%"REG_a", %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 32(%1, %%"REG_a", 2) \n\t"
PREFETCH" 32(%2, %%"REG_a") \n\t"
"xor %%"REG_a", %%"REG_a" \n\t"
"pcmpeqw %%mm7, %%mm7 \n\t"
"psrlw $8, %%mm7 \n\t" // FF,00,FF,00...
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 64(%0, %%"REG_a", 4) \n\t"
"movq (%0, %%"REG_a", 4), %%mm0 \n\t" // YUYV YUYV(0)
__asm__ volatile(
"xor %%"REG_a", %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 64(%0, %%"REG_a", 4) \n\t"
"movq (%0, %%"REG_a", 4), %%mm0 \n\t" // YUYV YUYV(0)
"xor %%"REG_a", %%"REG_a" \n\t"
"pcmpeqw %%mm7, %%mm7 \n\t"
"psrlw $8, %%mm7 \n\t" // FF,00,FF,00...
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 64(%0, %%"REG_a", 4) \n\t"
"movq (%0, %%"REG_a", 4), %%mm0 \n\t" // UYVY UYVY(0)
__asm__ volatile(
"xor %%"REG_a", %%"REG_a" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 64(%0, %%"REG_a", 4) \n\t"
"movq (%0, %%"REG_a", 4), %%mm0 \n\t" // YUYV YUYV(0)
"movq "MANGLE(ff_w1111)", %%mm5 \n\t"
"pxor %%mm7, %%mm7 \n\t"
"lea (%%"REG_a", %%"REG_a", 2), %%"REG_d" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 64(%0, %%"REG_d") \n\t"
"movd (%0, %%"REG_d"), %%mm0 \n\t"
"pxor %%mm7, %%mm7 \n\t"
"lea (%%"REG_a", %%"REG_a", 2), %%"REG_d" \n\t"
"add %%"REG_d", %%"REG_d" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
PREFETCH" 64(%0, %%"REG_d") \n\t"
PREFETCH" 64(%1, %%"REG_d") \n\t"
"movq %%mm3, %%mm4 \n\t"\
"lea " offset "(%0), %%"REG_d" \n\t"\
"mov (%%"REG_d"), %%"REG_S" \n\t"\
- ASMALIGN(4) /* FIXME Unroll? */\
+ ".p2align 4 \n\t" /* FIXME Unroll? */\
"1: \n\t"\
"movq 8(%%"REG_d"), %%mm0 \n\t" /* filterCoeff */\
"movq " x "(%%"REG_S", %%"REG_a", 2), %%mm2 \n\t" /* srcData */\
"pxor %%mm6, %%mm6 \n\t"\
"pxor %%mm7, %%mm7 \n\t"\
"mov (%%"REG_d"), %%"REG_S" \n\t"\
- ASMALIGN(4) \
+ ".p2align 4 \n\t"\
"1: \n\t"\
"movq " x "(%%"REG_S", %%"REG_a", 2), %%mm0 \n\t" /* srcData */\
"movq 8+" x "(%%"REG_S", %%"REG_a", 2), %%mm2 \n\t" /* srcData */\
#define YSCALEYUV2YV121 \
"mov %2, %%"REG_a" \n\t"\
- ASMALIGN(4) /* FIXME Unroll? */\
+ ".p2align 4 \n\t" /* FIXME Unroll? */\
"1: \n\t"\
"movq (%0, %%"REG_a", 2), %%mm0 \n\t"\
"movq 8(%0, %%"REG_a", 2), %%mm1 \n\t"\
"pcmpeqw %%mm7, %%mm7 \n\t"\
"psrlw $15, %%mm7 \n\t"\
"psllw $6, %%mm7 \n\t"\
- ASMALIGN(4) /* FIXME Unroll? */\
+ ".p2align 4 \n\t" /* FIXME Unroll? */\
"1: \n\t"\
"movq (%0, %%"REG_a", 2), %%mm0 \n\t"\
"movq 8(%0, %%"REG_a", 2), %%mm1 \n\t"\
#define YSCALEYUV2PACKEDX_UV \
__asm__ volatile(\
"xor %%"REG_a", %%"REG_a" \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"nop \n\t"\
"1: \n\t"\
"lea "CHR_MMX_FILTER_OFFSET"(%0), %%"REG_d" \n\t"\
"mov (%%"REG_d"), %%"REG_S" \n\t"\
"movq "VROUNDER_OFFSET"(%0), %%mm3 \n\t"\
"movq %%mm3, %%mm4 \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"2: \n\t"\
"movq 8(%%"REG_d"), %%mm0 \n\t" /* filterCoeff */\
"movq (%%"REG_S", %%"REG_a"), %%mm2 \n\t" /* UsrcData */\
"mov (%%"REG_d"), %%"REG_S" \n\t"\
"movq "VROUNDER_OFFSET"(%0), "#dst1" \n\t"\
"movq "#dst1", "#dst2" \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"2: \n\t"\
"movq 8(%%"REG_d"), "#coeff" \n\t" /* filterCoeff */\
"movq (%%"REG_S", %%"REG_a", 2), "#src1" \n\t" /* Y1srcData */\
#define YSCALEYUV2PACKEDX_ACCURATE_UV \
__asm__ volatile(\
"xor %%"REG_a", %%"REG_a" \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"nop \n\t"\
"1: \n\t"\
"lea "CHR_MMX_FILTER_OFFSET"(%0), %%"REG_d" \n\t"\
"pxor %%mm5, %%mm5 \n\t"\
"pxor %%mm6, %%mm6 \n\t"\
"pxor %%mm7, %%mm7 \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"2: \n\t"\
"movq (%%"REG_S", %%"REG_a"), %%mm0 \n\t" /* UsrcData */\
"movq "AV_STRINGIFY(VOF)"(%%"REG_S", %%"REG_a"), %%mm2 \n\t" /* VsrcData */\
"pxor %%mm5, %%mm5 \n\t"\
"pxor %%mm7, %%mm7 \n\t"\
"pxor %%mm6, %%mm6 \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"2: \n\t"\
"movq (%%"REG_S", %%"REG_a", 2), %%mm0 \n\t" /* Y1srcData */\
"movq 8(%%"REG_S", %%"REG_a", 2), %%mm2 \n\t" /* Y2srcData */\
"movq %%mm0, "CHR_MMX_FILTER_OFFSET"+8("#c") \n\t"\
"movq %%mm1, "LUM_MMX_FILTER_OFFSET"+8("#c") \n\t"\
"xor "#index", "#index" \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"1: \n\t"\
"movq (%2, "#index"), %%mm2 \n\t" /* uvbuf0[eax]*/\
"movq (%3, "#index"), %%mm3 \n\t" /* uvbuf1[eax]*/\
#define REAL_YSCALEYUV2RGB_UV(index, c) \
"xor "#index", "#index" \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"1: \n\t"\
"movq (%2, "#index"), %%mm2 \n\t" /* uvbuf0[eax]*/\
"movq (%3, "#index"), %%mm3 \n\t" /* uvbuf1[eax]*/\
#define REAL_YSCALEYUV2PACKED1(index, c) \
"xor "#index", "#index" \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"1: \n\t"\
"movq (%2, "#index"), %%mm3 \n\t" /* uvbuf0[eax]*/\
"movq "AV_STRINGIFY(VOF)"(%2, "#index"), %%mm4 \n\t" /* uvbuf0[eax+2048]*/\
#define REAL_YSCALEYUV2RGB1(index, c) \
"xor "#index", "#index" \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"1: \n\t"\
"movq (%2, "#index"), %%mm3 \n\t" /* uvbuf0[eax]*/\
"movq "AV_STRINGIFY(VOF)"(%2, "#index"), %%mm4 \n\t" /* uvbuf0[eax+2048]*/\
#define REAL_YSCALEYUV2PACKED1b(index, c) \
"xor "#index", "#index" \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"1: \n\t"\
"movq (%2, "#index"), %%mm2 \n\t" /* uvbuf0[eax]*/\
"movq (%3, "#index"), %%mm3 \n\t" /* uvbuf1[eax]*/\
// do vertical chrominance interpolation
#define REAL_YSCALEYUV2RGB1b(index, c) \
"xor "#index", "#index" \n\t"\
- ASMALIGN(4)\
+ ".p2align 4 \n\t"\
"1: \n\t"\
"movq (%2, "#index"), %%mm2 \n\t" /* uvbuf0[eax]*/\
"movq (%3, "#index"), %%mm3 \n\t" /* uvbuf1[eax]*/\
"pxor %%mm7, %%mm7 \n\t"
"push %%"REG_BP" \n\t" // we use 7 regs here ...
"mov %%"REG_a", %%"REG_BP" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movzwl (%2, %%"REG_BP"), %%eax \n\t"
"movzwl 2(%2, %%"REG_BP"), %%ebx \n\t"
"pxor %%mm7, %%mm7 \n\t"
"push %%"REG_BP" \n\t" // we use 7 regs here ...
"mov %%"REG_a", %%"REG_BP" \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movzwl (%2, %%"REG_BP"), %%eax \n\t"
"movzwl 2(%2, %%"REG_BP"), %%ebx \n\t"
dst-= counter/2;
__asm__ volatile(
"pxor %%mm7, %%mm7 \n\t"
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"mov %2, %%"REG_c" \n\t"
"movzwl (%%"REG_c", %0), %%eax \n\t"
"xor %%"REG_a", %%"REG_a" \n\t" // i
"xor %%"REG_d", %%"REG_d" \n\t" // xx
"xorl %%ecx, %%ecx \n\t" // xalpha
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"movzbl (%0, %%"REG_d"), %%edi \n\t" //src[xx]
"movzbl 1(%0, %%"REG_d"), %%esi \n\t" //src[xx+1]
"xor %%"REG_a", %%"REG_a" \n\t" // i
"xor %%"REG_d", %%"REG_d" \n\t" // xx
"xorl %%ecx, %%ecx \n\t" // xalpha
- ASMALIGN(4)
+ ".p2align 4 \n\t"
"1: \n\t"
"mov %0, %%"REG_S" \n\t"
"movzbl (%%"REG_S", %%"REG_d"), %%edi \n\t" //src[xx]