Altivec on non darwin systems patch by Romain Dolbeau

Originally committed as revision 1509 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/Makefile

@@ -86,7 +86,11 @@ OBJS += ps2/dsputil_mmi.o ps2/idct_mmi.o ps2/mpegvideo_mmi.o
 endif
 
 ifeq ($(TARGET_ALTIVEC),yes)
+ifeq ($(TARGET_OS),Darwin)
 CFLAGS += -faltivec
+else
+CFLAGS += -maltivec -mabi=altivec
+endif
 OBJS += ppc/dsputil_altivec.o ppc/mpegvideo_altivec.o ppc/idct_altivec.o \
         ppc/fft_altivec.o ppc/gmc_altivec.o
 endif

libavcodec/dsputil.h

@@ -224,6 +224,10 @@ void dsputil_init_alpha(DSPContext* c, unsigned mask);
 
 extern int mm_flags;
 
+#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
+#include <altivec.h>
+#endif
+
 #define __align8 __attribute__ ((aligned (16)))
 
 void dsputil_init_ppc(DSPContext* c, unsigned mask);

libavcodec/ppc/dsputil_altivec.h

@@ -63,7 +63,11 @@ extern int has_altivec(void);
 #define WORD_s2 0x18,0x19,0x1a,0x1b
 #define WORD_s3 0x1c,0x1d,0x1e,0x1f
 
+#ifdef CONFIG_DARWIN
 #define vcprm(a,b,c,d) (const vector unsigned char)(WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d)
+#else
+#define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d}
+#endif
 
 // vcprmle is used to keep the same index as in the SSE version.
 // it's the same as vcprm, with the index inversed
@@ -75,7 +79,12 @@ extern int has_altivec(void);
 #define FLOAT_n -1.
 #define FLOAT_p 1.
 
+
+#ifdef CONFIG_DARWIN
 #define vcii(a,b,c,d) (const vector float)(FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d)
+#else
+#define vcii(a,b,c,d) (const vector float){FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d}
+#endif
 
 #else /* HAVE_ALTIVEC */
 #ifdef ALTIVEC_USE_REFERENCE_C_CODE

libavcodec/ppc/dsputil_ppc.c

@@ -120,7 +120,6 @@ POWERPC_TBL_START_COUNT(powerpc_clear_blocks_dcbz32, 1);
       ((unsigned long*)blocks)[1] = 0L;
       ((unsigned long*)blocks)[2] = 0L;
       ((unsigned long*)blocks)[3] = 0L;
-      vec_st((vector short)(0), 0, blocks);
       i += 16;
     }
     for ( ; i < sizeof(DCTELEM)*6*64 ; i += 32) {
@@ -142,18 +141,17 @@ POWERPC_TBL_STOP_COUNT(powerpc_clear_blocks_dcbz32, 1);
 /* check dcbz report how many bytes are set to 0 by dcbz */
 long check_dcbz_effect(void)
 {
-  register char *fakedata = (char*)malloc(1024);
+  register char *fakedata = (char*)av_malloc(1024);
   register char *fakedata_middle;
   register long zero = 0;
   register long i = 0;
   long count = 0;
 
-  if (fakedata == NULL)
+  if (!fakedata)
   {
     return 0L;
   }
 
-
   fakedata_middle = (fakedata + 512);
 
   memset(fakedata, 0xFF, 1024);
@@ -166,7 +164,7 @@ long check_dcbz_effect(void)
       count++;
   }
 
-  free(fakedata);
+  av_free(fakedata);
   
   return count;
 }

libavcodec/ppc/fft_altivec.c

@@ -138,7 +138,11 @@ POWERPC_TBL_START_COUNT(altivec_fft_num, s->nbits >= 6);
 POWERPC_TBL_STOP_COUNT(altivec_fft_num, s->nbits >= 6);
 
 #else /* ALTIVEC_USE_REFERENCE_C_CODE */
+#ifdef CONFIG_DARWIN
     register const vector float vczero = (const vector float)(0.);
+#else
+    register const vector float vczero = (const vector float){0.,0.,0.,0.};
+#endif
     
     int ln = s->nbits;
     int	j, np, np2;

libavcodec/ppc/gmc_altivec.c

@@ -66,8 +66,8 @@ POWERPC_TBL_STOP_COUNT(altivec_gmc1_num, h == 8);
         (   x16)*(   y16), /* D */
         0, 0, 0, 0         /* padding */
       };
-    register const vector unsigned char vczero = (const vector unsigned char)(0);
-    register const vector unsigned short vcsr8 = (const vector unsigned short)(8);
+    register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
+    register const vector unsigned short vcsr8 = (const vector unsigned short)vec_splat_u16(8);
     register vector unsigned char dstv, dstv2, src_0, src_1, srcvA, srcvB, srcvC, srcvD;
     register vector unsigned short Av, Bv, Cv, Dv, rounderV, tempA, tempB, tempC, tempD;
     int i;

libavcodec/ppc/idct_altivec.c

@@ -151,6 +151,8 @@
     vx6 = vec_sra (vy6, shift);						\
     vx7 = vec_sra (vy7, shift);
 
+
+#ifdef CONFIG_DARWIN
 static const vector_s16_t constants[5] = {
     (vector_s16_t)(23170, 13573, 6518, 21895, -23170, -21895, 32, 31),
     (vector_s16_t)(16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725),
@@ -158,6 +160,16 @@ static const vector_s16_t constants[5] = {
     (vector_s16_t)(21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692),
     (vector_s16_t)(19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722)
 };
+#else
+// broken gcc
+static const vector_s16_t constants[5] = {
+    (vector_s16_t){23170, 13573, 6518, 21895, -23170, -21895, 32, 31},
+    (vector_s16_t){16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725},
+    (vector_s16_t){22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521},
+    (vector_s16_t){21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692},
+    (vector_s16_t){19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722}
+};
+#endif
 
 void idct_put_altivec(uint8_t* dest, int stride, vector_s16_t* block)
 {

libavcodec/ppc/mpegvideo_altivec.c

@@ -90,6 +90,13 @@ do { \
     vec = vec_splat(vec, 0); \
 }
 
+
+#ifdef CONFIG_DARWIN
+#define FOUROF(a) (a)
+#else
+// slower, for dumb non-apple GCC
+#define FOUROF(a) {a,a,a,a}
+#endif
 int dct_quantize_altivec(MpegEncContext* s, 
                         DCTELEM* data, int n,
                         int qscale, int* overflow)
@@ -97,7 +104,7 @@ int dct_quantize_altivec(MpegEncContext* s,
     int lastNonZero;
     vector float row0, row1, row2, row3, row4, row5, row6, row7;
     vector float alt0, alt1, alt2, alt3, alt4, alt5, alt6, alt7;
-    const vector float zero = (const vector float)(0.0f);
+    const vector float zero = (const vector float)FOUROF(0.);
 
     // Load the data into the row/alt vectors
     {
@@ -141,18 +148,18 @@ int dct_quantize_altivec(MpegEncContext* s,
 		// in the vector local variables, as floats, which we'll use during the
 		// quantize step...
     {
-        const vector float vec_0_298631336 = (vector float)(0.298631336f);
-        const vector float vec_0_390180644 = (vector float)(-0.390180644f);
-        const vector float vec_0_541196100 = (vector float)(0.541196100f);
-        const vector float vec_0_765366865 = (vector float)(0.765366865f);
-        const vector float vec_0_899976223 = (vector float)(-0.899976223f);
-        const vector float vec_1_175875602 = (vector float)(1.175875602f);
-        const vector float vec_1_501321110 = (vector float)(1.501321110f);
-        const vector float vec_1_847759065 = (vector float)(-1.847759065f);
-        const vector float vec_1_961570560 = (vector float)(-1.961570560f);
-        const vector float vec_2_053119869 = (vector float)(2.053119869f);
-        const vector float vec_2_562915447 = (vector float)(-2.562915447f);
-        const vector float vec_3_072711026 = (vector float)(3.072711026f);
+        const vector float vec_0_298631336 = (vector float)FOUROF(0.298631336f);
+        const vector float vec_0_390180644 = (vector float)FOUROF(-0.390180644f);
+        const vector float vec_0_541196100 = (vector float)FOUROF(0.541196100f);
+        const vector float vec_0_765366865 = (vector float)FOUROF(0.765366865f);
+        const vector float vec_0_899976223 = (vector float)FOUROF(-0.899976223f);
+        const vector float vec_1_175875602 = (vector float)FOUROF(1.175875602f);
+        const vector float vec_1_501321110 = (vector float)FOUROF(1.501321110f);
+        const vector float vec_1_847759065 = (vector float)FOUROF(-1.847759065f);
+        const vector float vec_1_961570560 = (vector float)FOUROF(-1.961570560f);
+        const vector float vec_2_053119869 = (vector float)FOUROF(2.053119869f);
+        const vector float vec_2_562915447 = (vector float)FOUROF(-2.562915447f);
+        const vector float vec_3_072711026 = (vector float)FOUROF(3.072711026f);
 
 
         int whichPass, whichHalf;
@@ -306,7 +313,7 @@ int dct_quantize_altivec(MpegEncContext* s,
 				// rounding when we convert to int, instead of flooring.)
         {
             vector signed int biasInt;
-            const vector float negOneFloat = (vector float)(-1.0f);
+            const vector float negOneFloat = (vector float)FOUROF(-1.0f);
             LOAD4(biasInt, biasAddr);
             bias = vec_ctf(biasInt, QUANT_BIAS_SHIFT);
             negBias = vec_madd(bias, negOneFloat, zero);
@@ -503,6 +510,7 @@ int dct_quantize_altivec(MpegEncContext* s,
 
     return lastNonZero;
 }
+#undef FOUROF
 
 /*
   AltiVec version of dct_unquantize_h263
@@ -551,7 +559,7 @@ POWERPC_TBL_START_COUNT(altivec_dct_unquantize_h263_num, 1);
     }
 #else /* ALTIVEC_USE_REFERENCE_C_CODE */
     {
-      register const vector short vczero = (const vector short)(0);
+      register const vector short vczero = (const vector short)vec_splat_s16(0);
       short __attribute__ ((aligned(16))) qmul8[] =
           {
             qmul, qmul, qmul, qmul,