summaryrefslogtreecommitdiff
path: root/libavcodec/x86/videodsp_init.c
diff options
context:
space:
mode:
authorRonald S. Bultje <rsbultje@gmail.com>2013-10-24 06:54:32 -0400
committerJanne Grunau <janne-libav@jannau.net>2014-01-30 15:30:01 +0100
commit51daafb02eaf96e0743a37ce95a7f5d02c1fa3c2 (patch)
tree2882850ccc912fab642661637d91814954d6e21a /libavcodec/x86/videodsp_init.c
parent64ba831da99c5526b21d510397b449742e92961e (diff)
x86: videodsp: Properly mark sse2 instructions in emulated_edge_mc as such.
Should fix crashes or corrupt output on pre-SSE2 CPUs when they were using SSE2-code (e.g. AMD Athlon XP 2400+ or Intel Pentium III) in hfix or hvar single-edge (left/right) extension functions. Signed-off-by: Janne Grunau <janne-libav@jannau.net>
Diffstat (limited to 'libavcodec/x86/videodsp_init.c')
-rw-r--r--libavcodec/x86/videodsp_init.c37
1 files changed, 27 insertions, 10 deletions
diff --git a/libavcodec/x86/videodsp_init.c b/libavcodec/x86/videodsp_init.c
index 79d980194c..8ee837096a 100644
--- a/libavcodec/x86/videodsp_init.c
+++ b/libavcodec/x86/videodsp_init.c
@@ -115,17 +115,17 @@ static emu_edge_hfix_func *hfixtbl_mmx[11] = {
};
#endif
extern emu_edge_hvar_func ff_emu_edge_hvar_mmx;
-extern emu_edge_hfix_func ff_emu_edge_hfix16_sse;
-extern emu_edge_hfix_func ff_emu_edge_hfix18_sse;
-extern emu_edge_hfix_func ff_emu_edge_hfix20_sse;
-extern emu_edge_hfix_func ff_emu_edge_hfix22_sse;
-static emu_edge_hfix_func *hfixtbl_sse[11] = {
+extern emu_edge_hfix_func ff_emu_edge_hfix16_sse2;
+extern emu_edge_hfix_func ff_emu_edge_hfix18_sse2;
+extern emu_edge_hfix_func ff_emu_edge_hfix20_sse2;
+extern emu_edge_hfix_func ff_emu_edge_hfix22_sse2;
+static emu_edge_hfix_func *hfixtbl_sse2[11] = {
ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx,
ff_emu_edge_hfix8_mmx, ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx,
- ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_sse, ff_emu_edge_hfix18_sse,
- ff_emu_edge_hfix20_sse, ff_emu_edge_hfix22_sse
+ ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_sse2, ff_emu_edge_hfix18_sse2,
+ ff_emu_edge_hfix20_sse2, ff_emu_edge_hfix22_sse2
};
-extern emu_edge_hvar_func ff_emu_edge_hvar_sse;
+extern emu_edge_hvar_func ff_emu_edge_hvar_sse2;
static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src,
ptrdiff_t dst_stride,
@@ -211,7 +211,6 @@ static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src,
src_x, src_y, w, h, vfixtbl_mmx, &ff_emu_edge_vvar_mmx,
hfixtbl_mmx, &ff_emu_edge_hvar_mmx);
}
-#endif
static av_noinline void emulated_edge_mc_sse(uint8_t * buf,const uint8_t *src,
ptrdiff_t buf_stride,
@@ -221,7 +220,20 @@ static av_noinline void emulated_edge_mc_sse(uint8_t * buf,const uint8_t *src,
{
emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse,
- hfixtbl_sse, &ff_emu_edge_hvar_sse);
+ hfixtbl_mmx, &ff_emu_edge_hvar_mmx);
+}
+#endif
+
+static av_noinline void emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src,
+ ptrdiff_t buf_stride,
+ ptrdiff_t src_stride,
+ int block_w, int block_h,
+ int src_x, int src_y, int w,
+ int h)
+{
+ emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h, src_x,
+ src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse,
+ hfixtbl_sse2, &ff_emu_edge_hvar_sse2);
}
#endif /* HAVE_YASM */
@@ -244,8 +256,13 @@ av_cold void ff_videodsp_init_x86(VideoDSPContext *ctx, int bpc)
if (EXTERNAL_MMXEXT(cpu_flags)) {
ctx->prefetch = ff_prefetch_mmxext;
}
+#if ARCH_X86_32
if (EXTERNAL_SSE(cpu_flags) && bpc <= 8) {
ctx->emulated_edge_mc = emulated_edge_mc_sse;
}
+#endif /* ARCH_X86_32 */
+ if (EXTERNAL_SSE2(cpu_flags) && bpc <= 8) {
+ ctx->emulated_edge_mc = emulated_edge_mc_sse2;
+ }
#endif /* HAVE_YASM */
}