* liboil/colorspace/argb_paint_i386.c:

* liboil/colorspace/ayuv2argb_i386.c:
* liboil/liboilfunction.c: (oil_class_optimize): disable functions
that fail test
* liboil/liboiltest.c: (oil_test_new), (check_zero),
(oil_test_check_impl), (init_parameter): Fix double-free bug, plus
other problems with testing from applications.
* liboil/dct/idct8x8_i386.c: pshufw apparently is not MMX
* liboil/simdpack/abs_i386.c: (abs_u16_s16_i386asm3),
(abs_u16_s16_mmx), (abs_u16_s16_mmxx): disable code that doesn\t

5 files changed, 35 insertions, 21 deletions

diff --git a/liboil/simdpack/abs_i386.c b/liboil/simdpack/abs_i386.c
index 7ecdb3c..626065e 100644
--- a/liboil/simdpack/abs_i386.c
+++ b/liboil/simdpack/abs_i386.c
@@ -34,6 +34,7 @@
 
 #define ABS(x) ((x)>0 ? (x) : -(x))
 
+#if 0
 static void
 abs_u16_s16_i386asm (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
 {
@@ -53,8 +54,10 @@ abs_u16_s16_i386asm (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
       ::"eax", "edx");
 }
 
-OIL_DEFINE_IMPL_ASM (abs_u16_s16_i386asm, abs_u16_s16);
+OIL_DEFINE_IMPL_FULL (abs_u16_s16_i386asm, abs_u16_s16, OIL_IMPL_FLAG_CMOV);
+#endif
 
+#if 0
 /* The previous function after running through uberopt */
 static void
 abs_u16_s16_i386asm_uber4 (uint16_t * dest, int dstr, int16_t * src,
@@ -76,8 +79,10 @@ abs_u16_s16_i386asm_uber4 (uint16_t * dest, int dstr, int16_t * src,
       :"+r" (src), "+r" (dest), "+r" (n)
       ::"eax", "edx");
 }
-OIL_DEFINE_IMPL_ASM (abs_u16_s16_i386asm_uber4, abs_u16_s16);
+OIL_DEFINE_IMPL_FULL (abs_u16_s16_i386asm_uber4, abs_u16_s16, OIL_IMPL_FLAG_CMOV);
+#endif
 
+#if 0
 static void
 abs_u16_s16_i386asm2 (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
 {
@@ -99,28 +104,26 @@ abs_u16_s16_i386asm2 (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
       "	popl	%%ebp			\n":"+D" (src), "+a" (dest), "+S" (n)
       ::"ecx", "edx");
 }
-
-OIL_DEFINE_IMPL_ASM (abs_u16_s16_i386asm2, abs_u16_s16);
+OIL_DEFINE_IMPL_FULL (abs_u16_s16_i386asm2, abs_u16_s16, OIL_IMPL_FLAG_CMOV);
+#endif
 
 static void
 abs_u16_s16_i386asm3 (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
 {
   __asm__ __volatile__ ("\n"
       "	.p2align 4,,15			\n"
-      "1:  movswl (%1), %%eax           \n"
-      "    mov %3, %%edx                \n"
-      "    add %%edx, %1                \n"
+      "1:  movsxw (%1), %%eax           \n"
+      "    add %3, %1                   \n"
       "    mov %%eax, %%edx             \n"
       "    sar $0xf, %%ax               \n"
       "    and %%edx, %%eax             \n"
       "    add %%eax, %%eax             \n"
       "    sub %%eax, %%edx             \n"
       "    mov %%dx, (%0)               \n"
-      "    mov %4, %%edx                \n"
-      "    add %%edx, %0                \n"
+      "    add %4, %0                   \n"
       "    decl %2                      \n"
       "    jne 1b                       \n"
-      : "+r" (src), "+r" (dest), "+m" (n)
+      : "+r" (dest), "+r" (src), "+m" (n)
       : "m" (dstr), "m" (sstr)
       : "eax", "edx");
 }
@@ -131,10 +134,11 @@ OIL_DEFINE_IMPL_ASM (abs_u16_s16_i386asm3, abs_u16_s16);
 static void
 abs_u16_s16_mmx (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
 {
-  const short p[] = { -32768, -32768, -32768, -32768,
-    32767, 32767, 32767, 32767
+  const int16_t p[][4] = {
+    { -32768, -32768, -32768, -32768 },
+    { 32767, 32767, 32767, 32767 }
   };
-  short tmp[4];
+  int16_t tmp[4];
 
   while (n & 3) {
     *dest = ABS (*src);
@@ -158,6 +162,7 @@ abs_u16_s16_mmx (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
     OIL_INCREMENT (src, sstr);
     __asm__ __volatile__ ("\n"
         "	movq	(%%eax), %%mm1		\n"
+        "	movq	%%mm1, %%mm0		\n"
         "	paddsw	%%mm2, %%mm0		\n"
         "	paddsw	%%mm3, %%mm1		\n"
         "	psubsw	%%mm2, %%mm0		\n"
@@ -180,6 +185,7 @@ abs_u16_s16_mmx (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
 
 OIL_DEFINE_IMPL_FULL (abs_u16_s16_mmx, abs_u16_s16, OIL_IMPL_FLAG_MMX);
 
+#if 0
 static void
 abs_u16_s16_mmxx (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
 {
@@ -222,8 +228,8 @@ abs_u16_s16_mmxx (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
       :"c" (p));
   asm volatile ("emms");
 }
-
 OIL_DEFINE_IMPL_FULL (abs_u16_s16_mmxx, abs_u16_s16, OIL_IMPL_FLAG_MMX);
+#endif
 
 static void
 abs_u16_s16_mmx2 (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
diff --git a/liboil/simdpack/clip_s32.c b/liboil/simdpack/clip_s32.c
index 001ae6f..464111c 100644
--- a/liboil/simdpack/clip_s32.c
+++ b/liboil/simdpack/clip_s32.c
@@ -35,19 +35,20 @@
 /* This is a suprisingly fast implementation of clipping
  * in straight C.  It would be difficult to do it faster in asm
  * without specialized opcodes.  However, this trick clips
- * the range min^(1<<31) to max^(1<<31) incorrectly.  So
- * it's limited to 31 bits. */
+ * the range min^(1<<31) to max^(1<<31) incorrectly with int32_t.
+ * Thus the use of int64_t. */
 
 static void
 clip_s32_fast (int32_t *dest, int dstr, int32_t *src, int sstr, int n,
     int32_t *low, int32_t *hi)
 {
 	int i;
-	int32_t x;
+	int64_t x;
 
 	for(i=0;i<n;i++){
-		x = src[i];
-		dest[i] = x - (((x-*low)>>31)&(x-*low)) + (((*hi-x)>>31)&(*hi-x));
+		x = OIL_GET(src,i*sstr,int32_t);
+		OIL_GET(dest,i*dstr,int32_t) = x - (((x-*low)>>31)&(x-*low))
+                  + (((*hi-x)>>31)&(*hi-x));
 	}
 }
 
diff --git a/liboil/simdpack/diffsquaresum_f64.c b/liboil/simdpack/diffsquaresum_f64.c
index f178902..633f90a 100644
--- a/liboil/simdpack/diffsquaresum_f64.c
+++ b/liboil/simdpack/diffsquaresum_f64.c
@@ -34,7 +34,7 @@
 #include <math.h>
 
 OIL_DEFINE_CLASS (diffsquaresum_f64,
-    "double *dest, double *src1, int sstr1, double *src2, int sstr2, int n");
+    "double *d_1, double *src1, int sstr1, double *src2, int sstr2, int n");
 
 static void
 diffsquaresum_f64_ref(double *dest, double *src1, int sstr1, double *src2,
@@ -108,7 +108,8 @@ diffsquaresum_f64_i10_unroll2(double *dest, double *src1, int sstr1, double *src
 }
 OIL_DEFINE_IMPL (diffsquaresum_f64_i10_unroll2, diffsquaresum_f64);
 
-static void diffsquaresum_f64_i10_unroll4(double *dest, double *src1, int sstr1,
+static void
+diffsquaresum_f64_i10_unroll4(double *dest, double *src1, int sstr1,
 	double *src2, int sstr2, int n)
 {
 	double sum0 = 0;
diff --git a/liboil/simdpack/scalaradd.c b/liboil/simdpack/scalaradd.c
index 4216a05..f5f8857 100644
--- a/liboil/simdpack/scalaradd.c
+++ b/liboil/simdpack/scalaradd.c
@@ -76,6 +76,9 @@ static void scalaradd_ ## type ## _unroll2(	\
     *dest = *src + *val;			\
     OIL_INCREMENT(dest,dstr);			\
     OIL_INCREMENT(src,sstr);			\
+    *dest = *src + *val;			\
+    OIL_INCREMENT(dest,dstr);			\
+    OIL_INCREMENT(src,sstr);			\
     n--;					\
   }						\
 }						\
diff --git a/liboil/simdpack/scalarmult.c b/liboil/simdpack/scalarmult.c
index 958b3c3..f4d8bb3 100644
--- a/liboil/simdpack/scalarmult.c
+++ b/liboil/simdpack/scalarmult.c
@@ -75,6 +75,9 @@ static void scalarmult_ ## type ## _unroll2(	\
     *dest = *src * *val;			\
     OIL_INCREMENT(dest,dstr);			\
     OIL_INCREMENT(src,sstr);			\
+    *dest = *src * *val;			\
+    OIL_INCREMENT(dest,dstr);			\
+    OIL_INCREMENT(src,sstr);			\
     n--;					\
   }						\
 }						\