1 files changed, 125 insertions, 6 deletions
diff --git a/apps/codecs/libwma/wmafixed.c b/apps/codecs/libwma/wmafixed.c
index 3c96700076..5569309145 100644
--- a/apps/codecs/libwma/wmafixed.c
+++ b/apps/codecs/libwma/wmafixed.c
@@ -63,20 +63,139 @@ fixed64 Fixed32To64(fixed32 x)
   return (fixed64)x;
 }
 
+
 /*
-    Not performance senstitive code here
+ * Helper functions for wma_window.
+ *
+ *
+ */
 
-*/
+#ifdef CPU_ARM
+inline void vector_fmul_add_add(fixed32 *dst, const fixed32 *data,
+                         const fixed32 *window, int n)
+{
+    /* Block sizes are always power of two */
+    asm volatile (
+        "0:"
+        "ldmia %[d]!, {r0, r1};"
+        "ldmia %[w]!, {r4, r5};"
+        /* consume the first data and window value so we can use those
+         * registers again */
+        "smull r8, r9, r0, r4;"
+        "ldmia %[dst], {r0, r4};"
+        "add   r0, r0, r9, lsl #1;"  /* *dst=*dst+(r9<<1)*/
+        "smull r8, r9, r1, r5;"
+        "add   r1, r4, r9, lsl #1;"
+        "stmia %[dst]!, {r0, r1};"
+        "subs  %[n], %[n], #2;"
+        "bne   0b;"
+        : [d] "+r" (data), [w] "+r" (window), [dst] "+r" (dst), [n] "+r" (n)
+        : : "r0", "r1", "r4", "r5", "r8", "r9", "memory", "cc");
+}
+
+inline void vector_fmul_reverse(fixed32 *dst, const fixed32 *src0, const fixed32 *src1,
+                         int len)
+{
+    /* Block sizes are always power of two */
+    asm volatile (
+        "add   %[s1], %[s1], %[n], lsl #2;"
+        "0:"
+        "ldmia %[s0]!, {r0, r1};"
+        "ldmdb %[s1]!, {r4, r5};"
+        "smull r8, r9, r0, r5;"
+        "mov   r0, r9, lsl #1;"
+        "smull r8, r9, r1, r4;"
+        "mov   r1, r9, lsl #1;"
+        "stmia %[dst]!, {r0, r1};"
+        "subs  %[n], %[n], #2;"
+        "bne   0b;"
+        : [s0] "+r" (src0), [s1] "+r" (src1), [dst] "+r" (dst), [n] "+r" (len)
+        : : "r0", "r1", "r4", "r5", "r8", "r9", "memory", "cc");
+}
 
+#elif defined(CPU_COLDFIRE)
 
-fixed64 fixmul64byfixed(fixed64 x, fixed32 y)
+inline void vector_fmul_add_add(fixed32 *dst, const fixed32 *data,
+                         const fixed32 *window, int n)
 {
+    /* Block sizes are always power of two. Smallest block is always way bigger
+     * than four too.*/
+    asm volatile (
+        "0:"
+        "movem.l (%[d]), %%d0-%%d3;"
+        "movem.l (%[w]), %%d4-%%d5/%%a0-%%a1;"
+        "mac.l %%d0, %%d4, %%acc0;"
+        "mac.l %%d1, %%d5, %%acc1;"
+        "mac.l %%d2, %%a0, %%acc2;"
+        "mac.l %%d3, %%a1, %%acc3;"
+        "lea.l (16, %[d]), %[d];"
+        "lea.l (16, %[w]), %[w];"
+        "movclr.l %%acc0, %%d0;"
+        "movclr.l %%acc1, %%d1;"
+        "movclr.l %%acc2, %%d2;"
+        "movclr.l %%acc3, %%d3;"
+        "movem.l (%[dst]), %%d4-%%d5/%%a0-%%a1;"
+        "add.l %%d4, %%d0;"
+        "add.l %%d5, %%d1;"
+        "add.l %%a0, %%d2;"
+        "add.l %%a1, %%d3;"
+        "movem.l %%d0-%%d3, (%[dst]);"
+        "lea.l (16, %[dst]), %[dst];"
+        "subq.l #4, %[n];"
+        "jne 0b;"
+        : [d] "+a" (data), [w] "+a" (window), [dst] "+a" (dst), [n] "+d" (n)
+        : : "d0", "d1", "d2", "d3", "d4", "d5", "a0", "a1", "memory", "cc");
+}
 
-    //return x * y;
-     return (x * y);
- // return (fixed64) fixmul32(Fixed32From64(x),y);
+inline void vector_fmul_reverse(fixed32 *dst, const fixed32 *src0, const fixed32 *src1,
+                         int len)
+{
+    /* Block sizes are always power of two. Smallest block is always way bigger
+     * than four too.*/
+    asm volatile (
+        "lea.l (-16, %[s1], %[n]*4), %[s1];"
+        "0:"
+        "movem.l (%[s0]), %%d0-%%d3;"
+        "movem.l (%[s1]), %%d4-%%d5/%%a0-%%a1;"
+        "mac.l %%d0, %%a1, %%acc0;"
+        "mac.l %%d1, %%a0, %%acc1;"
+        "mac.l %%d2, %%d5, %%acc2;"
+        "mac.l %%d3, %%d4, %%acc3;"
+        "lea.l (16, %[s0]), %[s0];"
+        "lea.l (-16, %[s1]), %[s1];"
+        "movclr.l %%acc0, %%d0;"
+        "movclr.l %%acc1, %%d1;"
+        "movclr.l %%acc2, %%d2;"
+        "movclr.l %%acc3, %%d3;"
+        "movem.l %%d0-%%d3, (%[dst]);"
+        "lea.l (16, %[dst]), %[dst];"
+        "subq.l #4, %[n];"
+        "jne 0b;"
+        : [s0] "+a" (src0), [s1] "+a" (src1), [dst] "+a" (dst), [n] "+d" (len)
+        : : "d0", "d1", "d2", "d3", "d4", "d5", "a0", "a1", "memory", "cc");
 }
 
+#else
+
+inline void vector_fmul_add_add(fixed32 *dst, const fixed32 *src0, const fixed32 *src1, int len){
+    int i;
+    for(i=0; i<len; i++)
+        dst[i] = fixmul32b(src0[i], src1[i]) + dst[i];
+}
+
+inline void vector_fmul_reverse(fixed32 *dst, const fixed32 *src0, const fixed32 *src1, int len){
+    int i;
+    src1 += len-1;
+    for(i=0; i<len; i++)
+        dst[i] = fixmul32b(src0[i], src1[-i]);
+}
+
+#endif
+
+/*
+    Not performance senstitive code here
+
+*/
 
 fixed32 fixdiv32(fixed32 x, fixed32 y)
 {