[FFmpeg-devel] [PATCH 2/5] swscale: preparation for use of AV_CPU_FLAG_*

[Janne Grunau]

add dsp_mask to swsContext and AVOptions to fill them
add function to convert sws_cpu_caps* to av_cpu_flag*
---
 libswscale/options.c          |   10 ++++++++++
 libswscale/swscale_internal.h |    1 +
 libswscale/utils.c            |   34 ++++++++++++++++++++++++++++++++--
 3 files changed, 43 insertions(+), 2 deletions(-)

diff --git a/libswscale/options.c b/libswscale/options.c
index 96e64c6..2d80941 100644
--- a/libswscale/options.c
+++ b/libswscale/options.c
@@ -19,6 +19,7 @@
  */
 
 #include "libavutil/avutil.h"
+#include "libavutil/cpu.h"
 #include "libavcodec/opt.h"
 #include "swscale.h"
 #include "swscale_internal.h"
@@ -68,6 +69,15 @@ static const AVOption options[] = {
     { "param0" , "scaler param 0" , OFFSET(param[0]) , FF_OPT_TYPE_DOUBLE, DEFAULT, INT_MIN, INT_MAX, VE },
     { "param1" , "scaler param 1" , OFFSET(param[1]) , FF_OPT_TYPE_DOUBLE, DEFAULT, INT_MIN, INT_MAX, VE },
 
+    { "sws_dsp_mask", "override cpu flags", OFFSET(dsp_mask), FF_OPT_TYPE_FLAGS, DEFAULT, 0, UINT_MAX, VE, "sws_dsp_mask" },
+    { "null", "no SIMD acceleration", 0, FF_OPT_TYPE_CONST, AV_CPU_FLAG_FORCE, INT_MIN, INT_MAX, VE, "sws_dsp_mask" },
+    { "mmx", "MMX SIMD acceleration", 0, FF_OPT_TYPE_CONST, AV_CPU_FLAG_FORCE|AV_CPU_FLAG_MMX, INT_MIN, INT_MAX, VE, "sws_dsp_mask" },
+    { "mmx2", "MMX2 SIMD acceleration", 0, FF_OPT_TYPE_CONST, AV_CPU_FLAG_FORCE|AV_CPU_FLAG_MMX2, INT_MIN, INT_MAX, VE, "sws_dsp_mask" },
+    { "sse2", "SSE2 SIMD acceleration", 0, FF_OPT_TYPE_CONST, AV_CPU_FLAG_FORCE|AV_CPU_FLAG_SSE2, INT_MIN, INT_MAX, VE, "sws_dsp_mask" },
+    { "3dnow", "3DNOW SIMD acceleration", 0, FF_OPT_TYPE_CONST, AV_CPU_FLAG_FORCE|AV_CPU_FLAG_3DNOW, INT_MIN, INT_MAX, VE, "sws_dsp_mask" },
+    { "altivec", "AltiVec SIMD acceleration", 0, FF_OPT_TYPE_CONST, AV_CPU_FLAG_FORCE|AV_CPU_FLAG_ALTIVEC, INT_MIN, UINT_MAX, VE, "sws_dsp_mask" },
+    { "bfin", "Blackfin SIMD acceleration", 0, FF_OPT_TYPE_CONST, AV_CPU_FLAG_FORCE|AV_CPU_FLAG_BFIN, INT_MIN, INT_MAX, VE, "sws_dsp_mask" },
+
     { NULL }
 };
 
diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
index d5df298..ad81a89 100644
--- a/libswscale/swscale_internal.h
+++ b/libswscale/swscale_internal.h
@@ -314,6 +314,7 @@ typedef struct SwsContext {
 
     int needs_hcscale; ///< Set if there are chroma planes to be converted.
 
+    unsigned dsp_mask; ///< Flags passed by the user to select optimizations
 } SwsContext;
 //FIXME check init (where 0)
 
diff --git a/libswscale/utils.c b/libswscale/utils.c
index b52a793..24dc1fc 100644
--- a/libswscale/utils.c
+++ b/libswscale/utils.c
@@ -43,6 +43,7 @@
 #include "libavutil/x86_cpu.h"
 #include "libavutil/avutil.h"
 #include "libavutil/bswap.h"
+#include "libavutil/cpu.h"
 #include "libavutil/pixdesc.h"
 
 unsigned swscale_version(void)
@@ -726,15 +727,21 @@ static int handle_jpeg(enum PixelFormat *format)
     }
 }
 
-static int update_flags_cpu(int flags)
+static int strip_cpu_flags(int flags)
 {
-#if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
     flags &= ~( SWS_CPU_CAPS_MMX
                |SWS_CPU_CAPS_MMX2
                |SWS_CPU_CAPS_3DNOW
                |SWS_CPU_CAPS_SSE2
                |SWS_CPU_CAPS_ALTIVEC
                |SWS_CPU_CAPS_BFIN);
+    return flags;
+}
+
+static int update_flags_cpu(int flags)
+{
+#if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
+    flags  = strip_cpu_flags(flags);
     flags |= ff_hardcodedcpuflags();
 #endif /* CONFIG_RUNTIME_CPUDETECT */
     return flags;
@@ -748,6 +755,26 @@ SwsContext *sws_alloc_context(void){
     return c;
 }
 
+static unsigned sws_cpu_caps2av_cpu_flags(int flags)
+{
+        unsigned cpuflags = 0;
+
+        if (ARCH_X86 && flags & SWS_CPU_CAPS_MMX)
+            cpuflags |= AV_CPU_FLAG_MMX;
+        if (ARCH_X86 && flags & SWS_CPU_CAPS_MMX2)
+            cpuflags |= AV_CPU_FLAG_MMX2;
+        if (ARCH_X86 && flags & SWS_CPU_CAPS_3DNOW)
+            cpuflags |= AV_CPU_FLAG_3DNOW;;
+        if (ARCH_X86 && flags & SWS_CPU_CAPS_SSE2)
+            cpuflags |= AV_CPU_FLAG_SSE2;
+        if (ARCH_PPC && flags & SWS_CPU_CAPS_ALTIVEC)
+            cpuflags |= AV_CPU_FLAG_ALTIVEC;
+        if (ARCH_BFIN && flags & SWS_CPU_CAPS_BFIN)
+            cpuflags |= AV_CPU_FLAG_BFIN;
+
+        return cpuflags;
+}
+
 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter){
     int i;
     int usesVFilter, usesHFilter;
@@ -758,9 +785,12 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter){
     int dstW= c->dstW;
     int dstH= c->dstH;
     int flags;
+    unsigned dsp_mask;
     enum PixelFormat srcFormat= c->srcFormat;
     enum PixelFormat dstFormat= c->dstFormat;
 
+    dsp_mask = c->dsp_mask = sws_cpu_caps2av_cpu_flags(update_flags_cpu(c->flags));
+
     flags= c->flags = update_flags_cpu(c->flags);
 #if ARCH_X86
     if (flags & SWS_CPU_CAPS_MMX)
-- 
1.7.3