Several tweaks and cleanups: * Use .rept instead of repeated macros for repeating blocks. * Use MUL (variant) instead of MLA (variant) in the first step of the ARM scalarproduct() if there's no loop. * Unroll ARM assembler functions to 32 where not already done, plus the generic scalarproduct().

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@19144 a1c6a512-1295-4272-9138-f99709370657

1 files changed, 45 insertions, 50 deletions

diff --git a/apps/codecs/demac/libdemac/vector_math32_armv4.h b/apps/codecs/demac/libdemac/vector_math32_armv4.h
index b729bd3a0a..89b24f2b06 100644
--- a/apps/codecs/demac/libdemac/vector_math32_armv4.h
+++ b/apps/codecs/demac/libdemac/vector_math32_armv4.h
@@ -30,27 +30,23 @@ static inline void vector_add(int32_t* v1, int32_t* v2)
     int cnt = ORDER>>5;
 #endif
 
-#define ADDBLOCK4                        \
-        "ldmia   %[v1],  {r0-r3}     \n" \
-        "ldmia   %[v2]!, {r4-r7}     \n" \
-        "add     r0, r0, r4          \n" \
-        "add     r1, r1, r5          \n" \
-        "add     r2, r2, r6          \n" \
-        "add     r3, r3, r7          \n" \
-        "stmia   %[v1]!, {r0-r3}     \n"
+#if ORDER > 16
+#define ADD_SUB_BLOCKS "8"
+#else
+#define ADD_SUB_BLOCKS "4"
+#endif
 
     asm volatile (
     "1:                              \n"
-        ADDBLOCK4
-        ADDBLOCK4
-        ADDBLOCK4
-        ADDBLOCK4
-#if ORDER > 16
-        ADDBLOCK4
-        ADDBLOCK4
-        ADDBLOCK4
-        ADDBLOCK4
-#endif
+        ".rept " ADD_SUB_BLOCKS     "\n"
+        "ldmia   %[v1],  {r0-r3}     \n"
+        "ldmia   %[v2]!, {r4-r7}     \n"
+        "add     r0, r0, r4          \n"
+        "add     r1, r1, r5          \n"
+        "add     r2, r2, r6          \n"
+        "add     r3, r3, r7          \n"
+        "stmia   %[v1]!, {r0-r3}     \n"
+        ".endr                       \n"
 #if ORDER > 32
         "subs    %[cnt], %[cnt], #1  \n"
         "bne     1b                  \n"
@@ -74,27 +70,17 @@ static inline void vector_sub(int32_t* v1, int32_t* v2)
     int cnt = ORDER>>5;
 #endif
 
-#define SUBBLOCK4                        \
-        "ldmia   %[v1],  {r0-r3}     \n" \
-        "ldmia   %[v2]!, {r4-r7}     \n" \
-        "sub     r0, r0, r4          \n" \
-        "sub     r1, r1, r5          \n" \
-        "sub     r2, r2, r6          \n" \
-        "sub     r3, r3, r7          \n" \
-        "stmia   %[v1]!, {r0-r3}     \n"
-
     asm volatile (
     "1:                              \n"
-        SUBBLOCK4
-        SUBBLOCK4
-        SUBBLOCK4
-        SUBBLOCK4
-#if ORDER > 16
-        SUBBLOCK4
-        SUBBLOCK4
-        SUBBLOCK4
-        SUBBLOCK4
-#endif
+        ".rept " ADD_SUB_BLOCKS     "\n"
+        "ldmia   %[v1],  {r0-r3}     \n"
+        "ldmia   %[v2]!, {r4-r7}     \n"
+        "sub     r0, r0, r4          \n"
+        "sub     r1, r1, r5          \n"
+        "sub     r2, r2, r6          \n"
+        "sub     r3, r3, r7          \n"
+        "stmia   %[v1]!, {r0-r3}     \n"
+        ".endr                       \n"
 #if ORDER > 32
         "subs    %[cnt], %[cnt], #1  \n"
         "bne     1b                  \n"
@@ -114,17 +100,24 @@ static inline void vector_sub(int32_t* v1, int32_t* v2)
 
 static inline int32_t scalarproduct(int32_t* v1, int32_t* v2)
 {
-    int res = 0;
+    int res;
 #if ORDER > 32
     int cnt = ORDER>>5;
 #endif
 
     asm volatile (
 #if ORDER > 16
+#if ORDER > 32
+        "mov     %[res], #0              \n"
+#endif
         "ldmia   %[v2]!, {r6-r7}         \n"
     "1:                                  \n"
         "ldmia   %[v1]!, {r0,r1,r3-r5}   \n"
+#if ORDER > 32
         "mla     %[res], r6, r0, %[res]  \n"
+#else
+        "mul     %[res], r6, r0          \n"
+#endif
         "mla     %[res], r7, r1, %[res]  \n"
         "ldmia   %[v2]!, {r0-r2,r6-r8}   \n"
         "mla     %[res], r0, r3, %[res]  \n"
@@ -177,19 +170,21 @@ static inline int32_t scalarproduct(int32_t* v1, int32_t* v2)
 #endif
 
 #else /* ORDER <= 16 */
-
-#define MLABLOCK4                            \
-        "ldmia   %[v1]!, {r0-r3}         \n" \
-        "ldmia   %[v2]!, {r4-r7}         \n" \
-        "mla     %[res], r4, r0, %[res]  \n" \
-        "mla     %[res], r5, r1, %[res]  \n" \
-        "mla     %[res], r6, r2, %[res]  \n" \
+        "ldmia   %[v1]!, {r0-r3}         \n"
+        "ldmia   %[v2]!, {r4-r7}         \n"
+        "mul     %[res], r4, r0          \n"
+        "mla     %[res], r5, r1, %[res]  \n"
+        "mla     %[res], r6, r2, %[res]  \n"
         "mla     %[res], r7, r3, %[res]  \n"
 
-        MLABLOCK4
-        MLABLOCK4
-        MLABLOCK4
-        MLABLOCK4
+        ".rept   3                       \n"
+        "ldmia   %[v1]!, {r0-r3}         \n"
+        "ldmia   %[v2]!, {r4-r7}         \n"
+        "mla     %[res], r4, r0, %[res]  \n"
+        "mla     %[res], r5, r1, %[res]  \n"
+        "mla     %[res], r6, r2, %[res]  \n"
+        "mla     %[res], r7, r3, %[res]  \n"
+        ".endr                           \n"
 #endif /* ORDER <= 16 */
         : /* outputs */
 #if ORDER > 32
@@ -197,7 +192,7 @@ static inline int32_t scalarproduct(int32_t* v1, int32_t* v2)
 #endif
         [v1] "+r"(v1),
         [v2] "+r"(v2),
-        [res]"+r"(res)
+        [res]"=r"(res)
         : /* inputs */
         : /* clobbers */
         "r0", "r1", "r2", "r3",