Merge ../mlt

[melted] / src / modules / core / composite_line_yuv_mmx.S

diff --git a/src/modules/core/composite_line_yuv_mmx.S b/src/modules/core/composite_line_yuv_mmx.S
index 7a5fb80..27af4dc 100644 (file)
--- a/src/modules/core/composite_line_yuv_mmx.S
+++ b/src/modules/core/composite_line_yuv_mmx.S
@@ -44,27 +44,34 @@ _composite_line_yuv_mmx:
        
 .loop:
 
-       movl $0xff, %ecx           # a = 255
+       movl $0xffff, %ecx           # a = 255
        cmpl $0, 20(%ebp)         # if alpha == NULL
        je .noalpha
        movl 20(%ebp), %edi       # a = alpha[ j ]
        movb (%edi,%edx), %cl
 .noalpha:
+       movl %ecx, -24(%ebp)      # save ecx
 
        movl 24(%ebp), %eax       # mix = weight
        cmpl $0, 28(%ebp)         # if luma == NULL
        je .noluma
        movl 28(%ebp), %edi       # mix = ...
        movl %edx, %ebx
-       #sall $1, %ebx
-       movw (%edi,%ebx), %ax # luma[ j*2 ]
-       cmpw %cx, %ax
+       sall $1, %ebx
+       movw (%edi,%ebx), %bx # luma[ j*2 ]
+       cmpl %ebx, %eax
        jl .luma0
-       movl %eax, %ebx
-       addl 32(%ebp), %ebx       # + softness
-       cmpw %bx, %cx
+       movl %ebx, %ecx
+       addl 32(%ebp), %ecx       # + softness
+       cmpl %ecx, %eax
        jge .luma1
-       /* TODO: linear interpolate between edges eax and ebx */
+       /* TODO: linear interpolate between edges */
+       subw %bx, %ax
+       sall $8, %eax
+       subw %bx, %cx
+       movl %edx, %ebx
+       divw %cx
+       movl %ebx, %edx
        jmp .noluma
 .luma0:
        movl $0, %eax
@@ -75,6 +82,7 @@ _composite_line_yuv_mmx:
        shrl $8, %eax
 
        movl %edx, %ebx           # edx will be destroyed by mulw
+       movl -24(%ebp), %ecx      # restore ecx
        mull %ecx                  # mix = mix * a...
        movl %ebx, %edx           # restore edx
        shrl $8, %eax             # >>8
@@ -137,8 +145,8 @@ _composite_line_yuv_mmx:
 
 /* for..next */
        addl $1, %edx             # j++
-       cmpl %edx, 16(%ebp)       # if ( j == width_src )
-       je .out
+       cmpl 16(%ebp), %edx       # if ( j == width_src )
+       jge .out
        
        addl $2, %esi
        addl $2, 12(%ebp)