rv40dsp x86: use only one register, for both increment and loop counter

Around 10 cycles faster for luma.
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>

libavcodec/x86/rv40dsp.asm

@@ -32,13 +32,14 @@ SECTION .text
 
 ; %1=5bits weights?, %2=dst %3=src1 %4=src3 %5=stride if sse2
 %macro RV40_WCORE  4-5
-    movh       m4, [%3 + 0]
-    movh       m5, [%4 + 0]
+    movh       m4, [%3 + r6 + 0]
+    movh       m5, [%4 + r6 + 0]
 %if %0 == 4
-%define OFFSET mmsize / 2
+%define OFFSET r6 + mmsize / 2
 %else
     ; 8x8 block and sse2, stride was provided
-%define OFFSET %5
+%define OFFSET r6
+    add        r6, r5
 %endif
     movh       m6, [%3 + OFFSET]
     movh       m7, [%4 + OFFSET]
@@ -99,10 +100,12 @@ SECTION .text
     packuswb   m4, m6
 %if %0 == 5
     ; Only called for 8x8 blocks and sse2
-    movh       [%2 +  0], m4
-    movhps     [%2 + %5], m4
+    sub        r6, r5
+    movh       [%2 + r6], m4
+    add        r6, r5
+    movhps     [%2 + r6], m4
 %else
-    mova       [%2], m4
+    mova       [%2 + r6], m4
 %endif
 %endmacro
 
@@ -115,26 +118,19 @@ SECTION .text
 %endif
 
     ; Prepare for next loop
-    add        r0, r5
-    add        r1, r5
-    add        r2, r5
+    add        r6, r5
 %else
 %ifidn %1, 8
     RV40_WCORE %2, r0, r1, r2, r5
     ; Prepare 2 next lines
-    lea        r0, [r0 + 2 * r5]
-    lea        r1, [r1 + 2 * r5]
-    lea        r2, [r2 + 2 * r5]
+    add        r6, r5
 %else
     RV40_WCORE %2, r0, r1, r2
     ; Prepare single next line
-    add        r0, r5
-    add        r1, r5
-    add        r2, r5
+    add        r6, r5
 %endif
 %endif
 
-    dec        r6
 %endmacro
 
 ; rv40_weight_func_%1(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, int stride)
@@ -145,7 +141,7 @@ SECTION .text
 ; Therefore, we check here whether they are multiples of 2^9 for
 ; those simplifications to occur.
 %macro RV40_WEIGHT  3
-cglobal rv40_weight_func_%1_%2, 6, 7, %3
+cglobal rv40_weight_func_%1_%2, 6, 7, 8
 %if cpuflag(ssse3)
     mova       m1, [shift_round]
 %else
@@ -153,11 +149,12 @@ cglobal rv40_weight_func_%1_%2, 6, 7, %3
 %endif
     pxor       m0, m0
     ; Set loop counter and increments
-%if mmsize == 8
-    mov        r6, %2
-%else
-    mov        r6, (%2 * %2) / mmsize
-%endif
+    mov        r6, r5
+    shl        r6, %3
+    add        r0, r6
+    add        r1, r6
+    add        r2, r6
+    neg        r6
 
     movd       m2, r3
     movd       m3, r4