[FFmpeg-cvslog] dsputil: Move MMX/SSE2-optimized IDCT bits to the x86 subdirectory

[Diego Biurrun]

ffmpeg | branch: master | Diego Biurrun <diego at biurrun.de> | Thu Jan 23 17:25:16 2014 +0100| [d2869aea0494d3a20d53d5034cd41dbb488eb133] | committer: Diego Biurrun

dsputil: Move MMX/SSE2-optimized IDCT bits to the x86 subdirectory

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=d2869aea0494d3a20d53d5034cd41dbb488eb133
---

 libavcodec/dsputil.c          |   27 +++++----------------------
 libavcodec/dsputil.h          |    2 ++
 libavcodec/x86/dsputil_init.c |   33 +++++++++++++++++++++++++++++++++
 3 files changed, 40 insertions(+), 22 deletions(-)

diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index 5d706bd..ca0c8ef 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -48,20 +48,6 @@ uint32_t ff_square_tab[512] = { 0, };
 #define BIT_DEPTH 8
 #include "dsputilenc_template.c"
 
-/* Input permutation for the simple_idct_mmx */
-static const uint8_t simple_mmx_permutation[64] = {
-    0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,
-    0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,
-    0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,
-    0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,
-    0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,
-    0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,
-    0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,
-    0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
-};
-
-static const uint8_t idct_sse2_row_perm[8] = { 0, 4, 1, 5, 2, 6, 3, 7 };
-
 av_cold void ff_init_scantable(uint8_t *permutation, ScanTable *st,
                                const uint8_t *src_scantable)
 {
@@ -88,6 +74,11 @@ av_cold void ff_init_scantable_permutation(uint8_t *idct_permutation,
 {
     int i;
 
+    if (ARCH_X86)
+        if (ff_init_scantable_permutation_x86(idct_permutation,
+                                              idct_permutation_type))
+            return;
+
     switch (idct_permutation_type) {
     case FF_NO_IDCT_PERM:
         for (i = 0; i < 64; i++)
@@ -97,10 +88,6 @@ av_cold void ff_init_scantable_permutation(uint8_t *idct_permutation,
         for (i = 0; i < 64; i++)
             idct_permutation[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);
         break;
-    case FF_SIMPLE_IDCT_PERM:
-        for (i = 0; i < 64; i++)
-            idct_permutation[i] = simple_mmx_permutation[i];
-        break;
     case FF_TRANSPOSE_IDCT_PERM:
         for (i = 0; i < 64; i++)
             idct_permutation[i] = ((i & 7) << 3) | (i >> 3);
@@ -109,10 +96,6 @@ av_cold void ff_init_scantable_permutation(uint8_t *idct_permutation,
         for (i = 0; i < 64; i++)
             idct_permutation[i] = (i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3);
         break;
-    case FF_SSE2_IDCT_PERM:
-        for (i = 0; i < 64; i++)
-            idct_permutation[i] = (i & 0x38) | idct_sse2_row_perm[i & 7];
-        break;
     default:
         av_log(NULL, AV_LOG_ERROR,
                "Internal error, IDCT permutation not set\n");
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index 1dbc784..b271dcc 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -57,6 +57,8 @@ void ff_init_scantable(uint8_t *permutation, ScanTable *st,
                        const uint8_t *src_scantable);
 void ff_init_scantable_permutation(uint8_t *idct_permutation,
                                    int idct_permutation_type);
+int ff_init_scantable_permutation_x86(uint8_t *idct_permutation,
+                                      int idct_permutation_type);
 
 /**
  * DSPContext.
diff --git a/libavcodec/x86/dsputil_init.c b/libavcodec/x86/dsputil_init.c
index 90e6271..74dab48 100644
--- a/libavcodec/x86/dsputil_init.c
+++ b/libavcodec/x86/dsputil_init.c
@@ -26,6 +26,39 @@
 #include "dsputil_x86.h"
 #include "idct_xvid.h"
 
+/* Input permutation for the simple_idct_mmx */
+static const uint8_t simple_mmx_permutation[64] = {
+    0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,
+    0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,
+    0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,
+    0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,
+    0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,
+    0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,
+    0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,
+    0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
+};
+
+static const uint8_t idct_sse2_row_perm[8] = { 0, 4, 1, 5, 2, 6, 3, 7 };
+
+av_cold int ff_init_scantable_permutation_x86(uint8_t *idct_permutation,
+                                              int idct_permutation_type)
+{
+    int i;
+
+    switch (idct_permutation_type) {
+    case FF_SIMPLE_IDCT_PERM:
+        for (i = 0; i < 64; i++)
+            idct_permutation[i] = simple_mmx_permutation[i];
+        return 1;
+    case FF_SSE2_IDCT_PERM:
+        for (i = 0; i < 64; i++)
+            idct_permutation[i] = (i & 0x38) | idct_sse2_row_perm[i & 7];
+        return 1;
+    }
+
+    return 0;
+}
+
 static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
                                      int cpu_flags, unsigned high_bit_depth)
 {