From bcf07a15a0aea23d2aa3eb71ea6ec7b3d21892ac Mon Sep 17 00:00:00 2001 From: Mans Rullgard Date: Wed, 3 Oct 2012 22:53:00 +0100 Subject: x86: dsputil: kill VLA in gmc_mmx() Instead of using an evil VLA, fall back to C version when edge emulation is needed. MPEG4 GMC is a rarely used fringe feature so the speed loss is an acceptable cost for safer code. Signed-off-by: Mans Rullgard --- libavcodec/x86/dsputil_mmx.c | 58 ++++++-------------------------------------- 1 file changed, 7 insertions(+), 51 deletions(-) (limited to 'libavcodec/x86') diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c index 121e7114c7..86a08cb06c 100644 --- a/libavcodec/x86/dsputil_mmx.c +++ b/libavcodec/x86/dsputil_mmx.c @@ -1909,15 +1909,10 @@ static av_noinline void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src, #if HAVE_INLINE_ASM -typedef void emulated_edge_mc_func(uint8_t *dst, const uint8_t *src, - int linesize, int block_w, int block_h, - int src_x, int src_y, int w, int h); - -static av_always_inline void gmc(uint8_t *dst, uint8_t *src, - int stride, int h, int ox, int oy, - int dxx, int dxy, int dyx, int dyy, - int shift, int r, int width, int height, - emulated_edge_mc_func *emu_edge_fn) +static void gmc_mmx(uint8_t *dst, uint8_t *src, + int stride, int h, int ox, int oy, + int dxx, int dxy, int dyx, int dyy, + int shift, int r, int width, int height) { const int w = 8; const int ix = ox >> (16 + shift); @@ -1932,7 +1927,6 @@ static av_always_inline void gmc(uint8_t *dst, uint8_t *src, const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys }; const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys }; const uint64_t shift2 = 2 * shift; - uint8_t edge_buf[(h + 1) * stride]; int x, y; const int dxw = (dxx - (1 << (16 + shift))) * (w - 1); @@ -1943,7 +1937,9 @@ static av_always_inline void gmc(uint8_t *dst, uint8_t *src, ((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) | (oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift) // uses more than 16 bits of subpel mv (only at huge resolution) - || (dxx | dxy | dyx | dyy) & 15) { + || (dxx | dxy | dyx | dyy) & 15 || + (unsigned)ix >= width - w || + (unsigned)iy >= height - h) { // FIXME could still use mmx for some of the rows ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height); @@ -1951,11 +1947,6 @@ static av_always_inline void gmc(uint8_t *dst, uint8_t *src, } src += ix + iy * stride; - if ((unsigned)ix >= width - w || - (unsigned)iy >= height - h) { - emu_edge_fn(edge_buf, src, stride, w + 1, h + 1, ix, iy, width, height); - src = edge_buf; - } __asm__ volatile ( "movd %0, %%mm6 \n\t" @@ -2034,36 +2025,6 @@ static av_always_inline void gmc(uint8_t *dst, uint8_t *src, } } -#if HAVE_YASM -#if ARCH_X86_32 -static void gmc_mmx(uint8_t *dst, uint8_t *src, - int stride, int h, int ox, int oy, - int dxx, int dxy, int dyx, int dyy, - int shift, int r, int width, int height) -{ - gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, - width, height, &emulated_edge_mc_mmx); -} -#endif -static void gmc_sse(uint8_t *dst, uint8_t *src, - int stride, int h, int ox, int oy, - int dxx, int dxy, int dyx, int dyy, - int shift, int r, int width, int height) -{ - gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, - width, height, &emulated_edge_mc_sse); -} -#else -static void gmc_mmx(uint8_t *dst, uint8_t *src, - int stride, int h, int ox, int oy, - int dxx, int dxy, int dyx, int dyy, - int shift, int r, int width, int height) -{ - gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, - width, height, &ff_emulated_edge_mc_8); -} -#endif - #define PREFETCH(name, op) \ static void name(void *mem, int stride, int h) \ { \ @@ -2469,9 +2430,7 @@ static void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx, int mm_flags) } } -#if ARCH_X86_32 || !HAVE_YASM c->gmc = gmc_mmx; -#endif c->add_bytes = add_bytes_mmx; @@ -2729,9 +2688,6 @@ static void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx, int mm_flags) if (!high_bit_depth) c->emulated_edge_mc = emulated_edge_mc_sse; -#if HAVE_INLINE_ASM - c->gmc = gmc_sse; -#endif #endif /* HAVE_YASM */ } -- cgit v1.2.3