x86/simple_idct: add explicit sse2 simple_idct_put/add versions.

These use the mmx IDCT, but sse2 put/add_pixels_clamped implementations.
This way we don't need to use the ff_put/add_pixels_clamped function
pointers.

libavcodec/x86/idctdsp_init.c

@@ -63,27 +63,41 @@ av_cold void ff_idctdsp_init_x86(IDCTDSPContext *c, AVCodecContext *avctx,
 {
     int cpu_flags = av_get_cpu_flags();
 
-    if (INLINE_MMX(cpu_flags)) {
-        if (!high_bit_depth &&
-            avctx->lowres == 0 &&
-            (avctx->idct_algo == FF_IDCT_AUTO ||
-             avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
-             avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
-                c->idct_put  = ff_simple_idct_put_mmx;
-                c->idct_add  = ff_simple_idct_add_mmx;
-                c->idct      = ff_simple_idct_mmx;
-                c->perm_type = FF_IDCT_PERM_SIMPLE;
-        }
-    }
     if (EXTERNAL_MMX(cpu_flags)) {
         c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
         c->put_pixels_clamped        = ff_put_pixels_clamped_mmx;
         c->add_pixels_clamped        = ff_add_pixels_clamped_mmx;
+
+        if (INLINE_MMX(cpu_flags)) {
+            if (!high_bit_depth &&
+                avctx->lowres == 0 &&
+                (avctx->idct_algo == FF_IDCT_AUTO ||
+                 avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
+                 avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
+                    c->idct_put  = ff_simple_idct_put_mmx;
+                    c->idct_add  = ff_simple_idct_add_mmx;
+                    c->idct      = ff_simple_idct_mmx;
+                    c->perm_type = FF_IDCT_PERM_SIMPLE;
+            }
+        }
     }
+
     if (EXTERNAL_SSE2(cpu_flags)) {
         c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_sse2;
         c->put_pixels_clamped        = ff_put_pixels_clamped_sse2;
         c->add_pixels_clamped        = ff_add_pixels_clamped_sse2;
+
+        if (INLINE_SSE2(cpu_flags)) {
+            if (!high_bit_depth &&
+                avctx->lowres == 0 &&
+                (avctx->idct_algo == FF_IDCT_AUTO ||
+                 avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
+                 avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
+                    c->idct_put  = ff_simple_idct_put_sse2;
+                    c->idct_add  = ff_simple_idct_add_sse2;
+                    c->perm_type = FF_IDCT_PERM_SIMPLE;
+            }
+        }
     }
 
     if (ARCH_X86_64 && avctx->lowres == 0) {

libavcodec/x86/simple_idct.c

@@ -24,6 +24,7 @@
 #include "libavutil/x86/asm.h"
 
 #include "libavcodec/idctdsp.h"
+#include "libavcodec/x86/idctdsp.h"
 
 #include "idctdsp.h"
 #include "simple_idct.h"
@@ -907,12 +908,22 @@ void ff_simple_idct_mmx(int16_t *block)
 void ff_simple_idct_put_mmx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
 {
     idct(block);
-    ff_put_pixels_clamped(block, dest, line_size);
+    ff_put_pixels_clamped_mmx(block, dest, line_size);
 }
 void ff_simple_idct_add_mmx(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
 {
     idct(block);
-    ff_add_pixels_clamped(block, dest, line_size);
+    ff_add_pixels_clamped_mmx(block, dest, line_size);
+}
+void ff_simple_idct_put_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
+{
+    idct(block);
+    ff_put_pixels_clamped_sse2(block, dest, line_size);
+}
+void ff_simple_idct_add_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
+{
+    idct(block);
+    ff_add_pixels_clamped_sse2(block, dest, line_size);
 }
 
 #endif /* HAVE_INLINE_ASM */

libavcodec/x86/simple_idct.h

@@ -26,6 +26,9 @@ void ff_simple_idct_mmx(int16_t *block);
 void ff_simple_idct_add_mmx(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
 void ff_simple_idct_put_mmx(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
 
+void ff_simple_idct_add_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
+void ff_simple_idct_put_sse2(uint8_t *dest, ptrdiff_t line_size, int16_t *block);
+
 void ff_simple_idct10_sse2(int16_t *block);
 void ff_simple_idct10_avx(int16_t *block);