Attempt to fix Win64 issues
This commit is contained in:
@@ -52,14 +52,14 @@ EXTN(jsimd_rgb_ycc_convert_sse2):
|
|||||||
collect_args
|
collect_args
|
||||||
push rbx
|
push rbx
|
||||||
|
|
||||||
mov rcx, r10
|
mov ecx, r10d
|
||||||
test rcx,rcx
|
test rcx,rcx
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
push rcx
|
push rcx
|
||||||
|
|
||||||
mov rsi, r12
|
mov rsi, r12
|
||||||
mov rcx, r13
|
mov ecx, r13d
|
||||||
mov rdi, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY]
|
mov rdi, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY]
|
||||||
mov rbx, JSAMPARRAY [rsi+1*SIZEOF_JSAMPARRAY]
|
mov rbx, JSAMPARRAY [rsi+1*SIZEOF_JSAMPARRAY]
|
||||||
mov rdx, JSAMPARRAY [rsi+2*SIZEOF_JSAMPARRAY]
|
mov rdx, JSAMPARRAY [rsi+2*SIZEOF_JSAMPARRAY]
|
||||||
@@ -70,7 +70,7 @@ EXTN(jsimd_rgb_ycc_convert_sse2):
|
|||||||
pop rcx
|
pop rcx
|
||||||
|
|
||||||
mov rsi, r11
|
mov rsi, r11
|
||||||
mov rax, r14
|
mov eax, r14d
|
||||||
test rax,rax
|
test rax,rax
|
||||||
jle near .return
|
jle near .return
|
||||||
.rowloop:
|
.rowloop:
|
||||||
|
|||||||
@@ -50,7 +50,7 @@ EXTN(jsimd_convsamp_float_sse2):
|
|||||||
packsswb xmm7,xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..)
|
packsswb xmm7,xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..)
|
||||||
|
|
||||||
mov rsi, r10
|
mov rsi, r10
|
||||||
mov rax, r11
|
mov eax, r11d
|
||||||
mov rdi, r12
|
mov rdi, r12
|
||||||
mov rcx, DCTSIZE/2
|
mov rcx, DCTSIZE/2
|
||||||
.convloop:
|
.convloop:
|
||||||
|
|||||||
@@ -50,7 +50,7 @@ EXTN(jsimd_convsamp_sse2):
|
|||||||
psllw xmm7,7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
|
psllw xmm7,7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}
|
||||||
|
|
||||||
mov rsi, r10
|
mov rsi, r10
|
||||||
mov rax, r11
|
mov eax, r11d
|
||||||
mov rdi, r12
|
mov rdi, r12
|
||||||
mov rcx, DCTSIZE/4
|
mov rcx, DCTSIZE/4
|
||||||
.convloop:
|
.convloop:
|
||||||
|
|||||||
@@ -49,11 +49,11 @@ EXTN(jsimd_h2v1_downsample_sse2):
|
|||||||
mov rbp,rsp
|
mov rbp,rsp
|
||||||
collect_args
|
collect_args
|
||||||
|
|
||||||
mov rcx, r13
|
mov ecx, r13d
|
||||||
shl rcx,3 ; imul rcx,DCTSIZE (rcx = output_cols)
|
shl rcx,3 ; imul rcx,DCTSIZE (rcx = output_cols)
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
mov rdx, r10
|
mov edx, r10d
|
||||||
|
|
||||||
; -- expand_right_edge
|
; -- expand_right_edge
|
||||||
|
|
||||||
@@ -62,7 +62,7 @@ EXTN(jsimd_h2v1_downsample_sse2):
|
|||||||
sub rcx,rdx
|
sub rcx,rdx
|
||||||
jle short .expand_end
|
jle short .expand_end
|
||||||
|
|
||||||
mov rax, r11
|
mov eax, r11d
|
||||||
test rax,rax
|
test rax,rax
|
||||||
jle short .expand_end
|
jle short .expand_end
|
||||||
|
|
||||||
@@ -90,7 +90,7 @@ EXTN(jsimd_h2v1_downsample_sse2):
|
|||||||
|
|
||||||
; -- h2v1_downsample
|
; -- h2v1_downsample
|
||||||
|
|
||||||
mov rax, r12 ; rowctr
|
mov eax, r12d ; rowctr
|
||||||
test eax,eax
|
test eax,eax
|
||||||
jle near .return
|
jle near .return
|
||||||
|
|
||||||
@@ -193,11 +193,11 @@ EXTN(jsimd_h2v2_downsample_sse2):
|
|||||||
mov rbp,rsp
|
mov rbp,rsp
|
||||||
collect_args
|
collect_args
|
||||||
|
|
||||||
mov rcx, r13
|
mov ecx, r13d
|
||||||
shl rcx,3 ; imul rcx,DCTSIZE (rcx = output_cols)
|
shl rcx,3 ; imul rcx,DCTSIZE (rcx = output_cols)
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
mov rdx, r10
|
mov edx, r10d
|
||||||
|
|
||||||
; -- expand_right_edge
|
; -- expand_right_edge
|
||||||
|
|
||||||
@@ -206,7 +206,7 @@ EXTN(jsimd_h2v2_downsample_sse2):
|
|||||||
sub rcx,rdx
|
sub rcx,rdx
|
||||||
jle short .expand_end
|
jle short .expand_end
|
||||||
|
|
||||||
mov rax, r11
|
mov eax, r11d
|
||||||
test rax,rax
|
test rax,rax
|
||||||
jle short .expand_end
|
jle short .expand_end
|
||||||
|
|
||||||
@@ -234,7 +234,7 @@ EXTN(jsimd_h2v2_downsample_sse2):
|
|||||||
|
|
||||||
; -- h2v2_downsample
|
; -- h2v2_downsample
|
||||||
|
|
||||||
mov rax, r12 ; rowctr
|
mov eax, r12d ; rowctr
|
||||||
test rax,rax
|
test rax,rax
|
||||||
jle near .return
|
jle near .return
|
||||||
|
|
||||||
|
|||||||
@@ -54,14 +54,14 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
|
|||||||
collect_args
|
collect_args
|
||||||
push rbx
|
push rbx
|
||||||
|
|
||||||
mov rcx, r10 ; num_cols
|
mov ecx, r10d ; num_cols
|
||||||
test rcx,rcx
|
test rcx,rcx
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
push rcx
|
push rcx
|
||||||
|
|
||||||
mov rdi, r11
|
mov rdi, r11
|
||||||
mov rcx, r12
|
mov ecx, r12d
|
||||||
mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
|
mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
|
||||||
mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
|
mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
|
||||||
mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
|
mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
|
||||||
@@ -72,7 +72,7 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
|
|||||||
pop rcx
|
pop rcx
|
||||||
|
|
||||||
mov rdi, r13
|
mov rdi, r13
|
||||||
mov rax, r14
|
mov eax, r14d
|
||||||
test rax,rax
|
test rax,rax
|
||||||
jle near .return
|
jle near .return
|
||||||
.rowloop:
|
.rowloop:
|
||||||
|
|||||||
@@ -54,14 +54,14 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
|
|||||||
collect_args
|
collect_args
|
||||||
push rbx
|
push rbx
|
||||||
|
|
||||||
mov rcx, r10 ; col
|
mov ecx, r10d ; col
|
||||||
test rcx,rcx
|
test rcx,rcx
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
push rcx
|
push rcx
|
||||||
|
|
||||||
mov rdi, r11
|
mov rdi, r11
|
||||||
mov rcx, r12
|
mov ecx, r12d
|
||||||
mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
|
mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
|
||||||
mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
|
mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
|
||||||
mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
|
mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
|
||||||
@@ -501,10 +501,10 @@ EXTN(jsimd_h2v2_merged_upsample_sse2):
|
|||||||
collect_args
|
collect_args
|
||||||
push rbx
|
push rbx
|
||||||
|
|
||||||
mov rax, r10
|
mov eax, r10d
|
||||||
|
|
||||||
mov rdi, r11
|
mov rdi, r11
|
||||||
mov rcx, r12
|
mov ecx, r12d
|
||||||
mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
|
mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY]
|
||||||
mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
|
mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY]
|
||||||
mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
|
mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY]
|
||||||
|
|||||||
@@ -67,11 +67,11 @@ EXTN(jsimd_h2v1_fancy_upsample_sse2):
|
|||||||
mov rbp,rsp
|
mov rbp,rsp
|
||||||
collect_args
|
collect_args
|
||||||
|
|
||||||
mov rax, r11 ; colctr
|
mov eax, r11d ; colctr
|
||||||
test rax,rax
|
test rax,rax
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
mov rcx, r10 ; rowctr
|
mov ecx, r10d ; rowctr
|
||||||
test rcx,rcx
|
test rcx,rcx
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
@@ -214,11 +214,11 @@ EXTN(jsimd_h2v2_fancy_upsample_sse2):
|
|||||||
collect_args
|
collect_args
|
||||||
push rbx
|
push rbx
|
||||||
|
|
||||||
mov rax, r11 ; colctr
|
mov eax, r11d ; colctr
|
||||||
test rax,rax
|
test rax,rax
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
mov rcx, r10 ; rowctr
|
mov ecx, r10d ; rowctr
|
||||||
test rcx,rcx
|
test rcx,rcx
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
@@ -506,12 +506,12 @@ EXTN(jsimd_h2v1_upsample_sse2):
|
|||||||
mov rbp,rsp
|
mov rbp,rsp
|
||||||
collect_args
|
collect_args
|
||||||
|
|
||||||
mov rdx, r11
|
mov edx, r11d
|
||||||
add rdx, byte (2*SIZEOF_XMMWORD)-1
|
add rdx, byte (2*SIZEOF_XMMWORD)-1
|
||||||
and rdx, byte -(2*SIZEOF_XMMWORD)
|
and rdx, byte -(2*SIZEOF_XMMWORD)
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
mov rcx, r10 ; rowctr
|
mov ecx, r10d ; rowctr
|
||||||
test rcx,rcx
|
test rcx,rcx
|
||||||
jz short .return
|
jz short .return
|
||||||
|
|
||||||
@@ -596,12 +596,12 @@ EXTN(jsimd_h2v2_upsample_sse2):
|
|||||||
collect_args
|
collect_args
|
||||||
push rbx
|
push rbx
|
||||||
|
|
||||||
mov rdx, r11
|
mov edx, r11d
|
||||||
add rdx, byte (2*SIZEOF_XMMWORD)-1
|
add rdx, byte (2*SIZEOF_XMMWORD)-1
|
||||||
and rdx, byte -(2*SIZEOF_XMMWORD)
|
and rdx, byte -(2*SIZEOF_XMMWORD)
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
mov rcx, r10 ; rowctr
|
mov ecx, r10d ; rowctr
|
||||||
test rcx,rcx
|
test rcx,rcx
|
||||||
jz near .return
|
jz near .return
|
||||||
|
|
||||||
|
|||||||
@@ -326,7 +326,7 @@ EXTN(jsimd_idct_float_sse2):
|
|||||||
mov rax, [original_rbp]
|
mov rax, [original_rbp]
|
||||||
lea rsi, [workspace] ; FAST_FLOAT * wsptr
|
lea rsi, [workspace] ; FAST_FLOAT * wsptr
|
||||||
mov rdi, r12 ; (JSAMPROW *)
|
mov rdi, r12 ; (JSAMPROW *)
|
||||||
mov rax, r13
|
mov eax, r13d
|
||||||
mov rcx, DCTSIZE/4 ; ctr
|
mov rcx, DCTSIZE/4 ; ctr
|
||||||
.rowloop:
|
.rowloop:
|
||||||
|
|
||||||
|
|||||||
@@ -323,7 +323,7 @@ EXTN(jsimd_idct_ifast_sse2):
|
|||||||
|
|
||||||
mov rax, [original_rbp]
|
mov rax, [original_rbp]
|
||||||
mov rdi, r12 ; (JSAMPROW *)
|
mov rdi, r12 ; (JSAMPROW *)
|
||||||
mov rax, r13
|
mov eax, r13d
|
||||||
|
|
||||||
; -- Even part
|
; -- Even part
|
||||||
|
|
||||||
|
|||||||
@@ -515,7 +515,7 @@ EXTN(jsimd_idct_islow_sse2):
|
|||||||
|
|
||||||
mov rax, [original_rbp]
|
mov rax, [original_rbp]
|
||||||
mov rdi, r12 ; (JSAMPROW *)
|
mov rdi, r12 ; (JSAMPROW *)
|
||||||
mov rax, r13
|
mov eax, r13d
|
||||||
|
|
||||||
; -- Even part
|
; -- Even part
|
||||||
|
|
||||||
|
|||||||
@@ -312,7 +312,7 @@ EXTN(jsimd_idct_4x4_sse2):
|
|||||||
|
|
||||||
mov rax, [original_rbp]
|
mov rax, [original_rbp]
|
||||||
mov rdi, r12 ; (JSAMPROW *)
|
mov rdi, r12 ; (JSAMPROW *)
|
||||||
mov rax, r13
|
mov eax, r13d
|
||||||
|
|
||||||
; -- Even part
|
; -- Even part
|
||||||
|
|
||||||
@@ -521,7 +521,7 @@ EXTN(jsimd_idct_2x2_sse2):
|
|||||||
; ---- Pass 2: process rows, store into output array.
|
; ---- Pass 2: process rows, store into output array.
|
||||||
|
|
||||||
mov rdi, r12 ; (JSAMPROW *)
|
mov rdi, r12 ; (JSAMPROW *)
|
||||||
mov rax, r13
|
mov eax, r13d
|
||||||
|
|
||||||
; | input:| result:|
|
; | input:| result:|
|
||||||
; | A0 B0 | |
|
; | A0 B0 | |
|
||||||
|
|||||||
Reference in New Issue
Block a user