VP8: armv6 optimizations.

From 52.503s (~40fps) to 27.973sec (~80fps) decoding of 480p sintel
trailer, i.e. a ~2x speedup overall, on a Nexus S.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/arm/Makefile

@@ -11,7 +11,8 @@ ARMV6-OBJS-$(CONFIG_MPEGAUDIODSP)      += arm/mpegaudiodsp_fixed_armv6.o
 OBJS-$(CONFIG_VP5_DECODER)             += arm/vp56dsp_init_arm.o
 OBJS-$(CONFIG_VP6_DECODER)             += arm/vp56dsp_init_arm.o
 OBJS-$(CONFIG_VP8_DECODER)             += arm/vp8dsp_init_arm.o
-ARMV6-OBJS-$(CONFIG_VP8_DECODER)       += arm/vp8_armv6.o
+ARMV6-OBJS-$(CONFIG_VP8_DECODER)       += arm/vp8_armv6.o               \
+                                          arm/vp8dsp_armv6.o
 
 OBJS-$(CONFIG_H264DSP)                 += arm/h264dsp_init_arm.o
 OBJS-$(CONFIG_H264PRED)                += arm/h264pred_init_arm.o

libavcodec/arm/asm.S

@@ -97,6 +97,12 @@ T       add             \rn, \rn, \rm
 T       ldr             \rt, [\rn]
 .endm
 
+.macro  ldr_dpren       rt,  rn,  rm:vararg
+A       ldr             \rt, [\rn, -\rm]
+T       sub             \rt, \rn, \rm
+T       ldr             \rt, [\rt]
+.endm
+
 .macro  ldr_post        rt,  rn,  rm:vararg
 A       ldr             \rt, [\rn], \rm
 T       ldr             \rt, [\rn]
@@ -133,6 +139,12 @@ T       ldrh            \rt, [\rn]
 T       add             \rn, \rn, \rm
 .endm
 
+.macro  ldrb_post       rt,  rn,  rm
+A       ldrb            \rt, [\rn], \rm
+T       ldrb            \rt, [\rn]
+T       add             \rn, \rn, \rm
+.endm
+
 .macro  str_post       rt,  rn,  rm:vararg
 A       str             \rt, [\rn], \rm
 T       str             \rt, [\rn]

libavcodec/arm/vp8dsp_neon.S

@@ -76,18 +76,6 @@ function ff_vp8_luma_dc_wht_neon, export=1
         bx              lr
 endfunc
 
-function ff_vp8_luma_dc_wht_dc_neon, export=1
-        ldrsh           r2,  [r1]
-        mov             r3,  #0
-        add             r2,  r2,  #3
-        strh            r3,  [r1]
-        asr             r2,  r2,  #3
-    .rept 16
-        strh            r2,  [r0], #32
-    .endr
-        bx              lr
-endfunc
-
 function ff_vp8_idct_add_neon, export=1
         vld1.16         {q0-q1},  [r1,:128]
         movw            r3,  #20091
@@ -741,23 +729,6 @@ function ff_put_vp8_pixels8_neon, export=1
         bx              lr
 endfunc
 
-function ff_put_vp8_pixels4_neon, export=1
-        ldr             r12, [sp, #0]           @ h
-        push            {r4-r6,lr}
-1:
-        subs            r12, r12, #4
-        ldr_post        r4,  r2,  r3
-        ldr_post        r5,  r2,  r3
-        ldr_post        r6,  r2,  r3
-        ldr_post        lr,  r2,  r3
-        str_post        r4,  r0,  r1
-        str_post        r5,  r0,  r1
-        str_post        r6,  r0,  r1
-        str_post        lr,  r0,  r1
-        bgt             1b
-        pop             {r4-r6,pc}
-endfunc
-
 /* 4/6-tap 8th-pel MC */
 
 .macro  vp8_epel8_h6    d,   a,   b