diff armv4l/mpegvideo_armv5te.c @ 8197:06acc3ab4bdc libavcodec

ARM: move dct_unquantize_h263_*_armv5te asm to separate file

--- a/armv4l/mpegvideo_armv5te.c	Sat Nov 22 16:36:50 2008 +0000
+++ b/armv4l/mpegvideo_armv5te.c	Sun Nov 23 19:11:44 2008 +0000
@@ -23,6 +23,8 @@
 #include "libavcodec/dsputil.h"
 #include "libavcodec/mpegvideo.h"
 
+extern void ff_dct_unquantize_h263_armv5te(DCTELEM *block, int qmul, int qadd,
+                                           int count);
 
 #ifdef ENABLE_ARM_TESTS
 /**
@@ -47,108 +49,6 @@
 }
 #endif
 
-/* GCC 3.1 or higher is required to support symbolic names in assembly code */
-#if (__GNUC__ > 3) || ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 1))
-
-/**
- * Special optimized version of dct_unquantize_h263_helper_c, it requires the block
- * to be at least 8 bytes aligned, and may process more elements than requested.
- * But it is guaranteed to never process more than 64 elements provided that
- * xxcount argument is <= 64, so it is safe. This macro is optimized for a common
- * distribution of values for nCoeffs (they are mostly multiple of 8 plus one or
- * two extra elements). So this macro processes data as 8 elements per loop iteration
- * and contains optional 2 elements processing in the end.
- *
- * Inner loop should take 6 cycles per element on arm926ej-s (Nokia 770)
- */
-#define dct_unquantize_h263_special_helper_armv5te(xxblock, xxqmul, xxqadd, xxcount) \
-({ DCTELEM *xblock = xxblock; \
-   int xqmul = xxqmul, xqadd = xxqadd, xcount = xxcount, xtmp; \
-   int xdata1, xdata2; \
-__asm__ volatile( \
-        "subs %[count], %[count], #2       \n\t" \
-        "ble 2f                            \n\t" \
-        "ldrd r4, [%[block], #0]           \n\t" \
-        "1:                                \n\t" \
-        "ldrd r6, [%[block], #8]           \n\t" \
-\
-        "rsbs %[data1], %[zero], r4, asr #16 \n\t" \
-        "addgt %[data1], %[qadd], #0       \n\t" \
-        "rsblt %[data1], %[qadd], #0       \n\t" \
-        "smlatbne %[data1], r4, %[qmul], %[data1] \n\t" \
-\
-        "rsbs %[data2], %[zero], r5, asr #16 \n\t" \
-        "addgt %[data2], %[qadd], #0       \n\t" \
-        "rsblt %[data2], %[qadd], #0       \n\t" \
-        "smlatbne %[data2], r5, %[qmul], %[data2] \n\t" \
-\
-        "rsbs %[tmp], %[zero], r4, asl #16 \n\t" \
-        "addgt %[tmp], %[qadd], #0         \n\t" \
-        "rsblt %[tmp], %[qadd], #0         \n\t" \
-        "smlabbne r4, r4, %[qmul], %[tmp]  \n\t" \
-\
-        "rsbs %[tmp], %[zero], r5, asl #16 \n\t" \
-        "addgt %[tmp], %[qadd], #0         \n\t" \
-        "rsblt %[tmp], %[qadd], #0         \n\t" \
-        "smlabbne r5, r5, %[qmul], %[tmp]  \n\t" \
-\
-        "strh r4, [%[block]], #2           \n\t" \
-        "strh %[data1], [%[block]], #2     \n\t" \
-        "strh r5, [%[block]], #2           \n\t" \
-        "strh %[data2], [%[block]], #2     \n\t" \
-\
-        "rsbs %[data1], %[zero], r6, asr #16 \n\t" \
-        "addgt %[data1], %[qadd], #0        \n\t" \
-        "rsblt %[data1], %[qadd], #0        \n\t" \
-        "smlatbne %[data1], r6, %[qmul], %[data1] \n\t" \
-\
-        "rsbs %[data2], %[zero], r7, asr #16 \n\t" \
-        "addgt %[data2], %[qadd], #0        \n\t" \
-        "rsblt %[data2], %[qadd], #0        \n\t" \
-        "smlatbne %[data2], r7, %[qmul], %[data2] \n\t" \
-\
-        "rsbs %[tmp], %[zero], r6, asl #16  \n\t" \
-        "addgt %[tmp], %[qadd], #0          \n\t" \
-        "rsblt %[tmp], %[qadd], #0          \n\t" \
-        "smlabbne r6, r6, %[qmul], %[tmp]   \n\t" \
-\
-        "rsbs %[tmp], %[zero], r7, asl #16  \n\t" \
-        "addgt %[tmp], %[qadd], #0          \n\t" \
-        "rsblt %[tmp], %[qadd], #0          \n\t" \
-        "smlabbne r7, r7, %[qmul], %[tmp]   \n\t" \
-\
-        "strh r6, [%[block]], #2            \n\t" \
-        "strh %[data1], [%[block]], #2      \n\t" \
-        "strh r7, [%[block]], #2            \n\t" \
-        "strh %[data2], [%[block]], #2      \n\t" \
-\
-        "subs %[count], %[count], #8        \n\t" \
-        "ldrgtd r4, [%[block], #0]          \n\t" /* load data early to avoid load/use pipeline stall */ \
-        "bgt 1b                             \n\t" \
-\
-        "adds %[count], %[count], #2        \n\t" \
-        "ble  3f                            \n\t" \
-        "2:                                 \n\t" \
-        "ldrsh %[data1], [%[block], #0]     \n\t" \
-        "ldrsh %[data2], [%[block], #2]     \n\t" \
-        "mov  %[tmp], %[qadd]               \n\t" \
-        "cmp  %[data1], #0                  \n\t" \
-        "rsblt %[tmp], %[qadd], #0          \n\t" \
-        "smlabbne %[data1], %[data1], %[qmul], %[tmp] \n\t" \
-        "mov  %[tmp], %[qadd]               \n\t" \
-        "cmp  %[data2], #0                  \n\t" \
-        "rsblt %[tmp], %[qadd], #0          \n\t" \
-        "smlabbne %[data2], %[data2], %[qmul], %[tmp] \n\t" \
-        "strh %[data1], [%[block]], #2      \n\t" \
-        "strh %[data2], [%[block]], #2      \n\t" \
-        "3:                                 \n\t" \
-        : [block] "+&r" (xblock), [count] "+&r" (xcount), [tmp] "=&r" (xtmp), \
-          [data1] "=&r" (xdata1), [data2] "=&r" (xdata2)  \
-        : [qmul] "r" (xqmul), [qadd] "r" (xqadd), [zero] "r" (0) \
-        : "r4", "r5", "r6", "r7", "cc", "memory" \
-); \
-})
-
 static void dct_unquantize_h263_intra_armv5te(MpegEncContext *s,
                                   DCTELEM *block, int n, int qscale)
 {
@@ -174,7 +74,7 @@
     else
         nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
 
-    dct_unquantize_h263_special_helper_armv5te(block, qmul, qadd, nCoeffs + 1);
+    ff_dct_unquantize_h263_armv5te(block, qmul, qadd, nCoeffs + 1);
     block[0] = level;
 }
 
@@ -191,17 +91,11 @@
 
     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
 
-    dct_unquantize_h263_special_helper_armv5te(block, qmul, qadd, nCoeffs + 1);
+    ff_dct_unquantize_h263_armv5te(block, qmul, qadd, nCoeffs + 1);
 }
 
-#define HAVE_DCT_UNQUANTIZE_H263_ARMV5TE_OPTIMIZED
-
-#endif
-
 void MPV_common_init_armv5te(MpegEncContext *s)
 {
-#ifdef HAVE_DCT_UNQUANTIZE_H263_ARMV5TE_OPTIMIZED
     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_armv5te;
     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_armv5te;
-#endif
 }