为了学习ARM程序集，我编写了一个简单的测试项目，使用内联汇编和NEON指令执行图像缩小。你可以在这里看到它：
https://github.com/rmaz/NEON-Image-Downscaling
经过一番努力，我设法让它工作，快乐的日子。除了它只适用于小于-O2的优化级别。我已经看了一下生成的ASM，但我看不出为什么会出现这种情况有任何明显的原因。谁能提供任何见解？这是负责内联汇编部分的函数：static void inline resizeRow(uint32_t *dst, uint32_t *src, uint32_t pixelsPerRow)
{
const uint32_t * rowB = src + pixelsPerRow;
// force the number of pixels per row to a mutliple of 8
pixelsPerRow = 8 * (pixelsPerRow / 8);
__asm__ volatile("Lresizeloop: \n" // start loop
"vld1.32 {d0-d3}, [%1]! \n" // load 8 pixels from the top row
"vld1.32 {d4-d7}, [%2]! \n" // load 8 pixels from the bottom row
"vhadd.u8 q0, q0, q2 \n" // average the pixels vertically
"vhadd.u8 q1, q1, q3 \n"
"vtrn.32 q0, q2 \n" // transpose to put the horizontally adjacent pixels in different registers
"vtrn.32 q1, q3 \n"
"vhadd.u8 q0, q0, q2 \n" // average the pixels horizontally
"vhadd.u8 q1, q1, q3 \n"
"vtrn.32 d0, d1 \n" // fill the registers with pixels
"vtrn.32 d2, d3 \n"
"vswp d1, d2 \n"
"vst1.64 {d0-d1}, [%0]! \n" // store the result
"subs %3, %3, #8 \n" // subtract 8 from the pixel count
"bne Lresizeloop \n" // repeat until the row is complete
: "=r"(dst), "=r"(src), "=r"(rowB), "=r"(pixelsPerRow)
: "0"(dst), "1"(src), "2"(rowB), "3"(pixelsPerRow)
: "q0", "q1", "q2", "q3"
);
}
O1周围功能和循环的功能生成输出如下：
.align 2
.code 16 @ @"\01-[BDPViewController downscaleImageNeon:]"
.thumb_func "-[BDPViewController downscaleImageNeon:]"
"-[BDPViewController downscaleImageNeon:]":
.cfi_startproc
Lfunc_begin4:
.loc 1 86 0 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:86:0
@ BB#0:
.loc 1 86 1 prologue_end @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:86:1
push {r4, r5, r6, r7, lr}
add r7, sp, #12
push.w {r8, r10, r11}
sub sp, #20
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:self <- R0+0
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:_cmd <- R1+0
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:image <- R2+0
.loc 1 88 20 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:88:20
Ltmp41:
movw r0, :lower16:(L_OBJC_SELECTOR_REFERENCES_2-(LPC4_0+4))
Ltmp42:
mov r6, r2
Ltmp43:
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:image <- R6+0
movt r0, :upper16:(L_OBJC_SELECTOR_REFERENCES_2-(LPC4_0+4))
LPC4_0:
add r0, pc
ldr.w r11, [r0]
mov r0, r6
blx _objc_retain
mov r4, r0
mov r0, r6
mov r1, r11
Ltmp44:
blx _objc_msgSend
blx _CGImageGetWidth
mov r5, r0
Ltmp45:
@DEBUG_VALUE: width <- R5+0
.loc 1 89 21 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:89:21
mov r0, r6
mov r1, r11
str r5, [sp, #16] @ 4-byte Spill
blx _objc_msgSend
blx _CGImageGetHeight
mov r10, r0
Ltmp46:
@DEBUG_VALUE: height <- R10+0
.loc 1 90 26 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:90:26
mov r0, r6
mov r1, r11
blx _objc_msgSend
blx _CGImageGetBytesPerRow
str r0, [sp, #12] @ 4-byte Spill
Ltmp47:
@DEBUG_VALUE: bytesPerRow <- [sp+#12]+#0
.loc 1 91 35 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:91:35
mov r0, r6
mov r1, r11
blx _objc_msgSend
blx _CGImageGetAlphaInfo
str r0, [sp, #4] @ 4-byte Spill
Ltmp48:
@DEBUG_VALUE: imageAlpha <- [sp+#4]+#0
.loc 1 94 45 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:94:45
mov r0, r6
mov r1, r11
blx _objc_msgSend
mov r6, r0
Ltmp49:
mov r0, r4
blx _objc_release
mov r0, r6
.loc 1 98 29 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:98:29
mul r8, r10, r5
Ltmp50:
@DEBUG_VALUE: width <- [sp+#16]+#0
.loc 1 94 45 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:94:45
blx _CGImageGetDataProvider
blx _CGDataProviderCopyData
Ltmp51:
@DEBUG_VALUE: data <- R0+0
str r0, [sp, #8] @ 4-byte Spill
Ltmp52:
@DEBUG_VALUE: data <- [sp+#8]+#0
.loc 1 95 29 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:95:29
blx _CFDataGetBytePtr
mov r4, r0
Ltmp53:
@DEBUG_VALUE: buffer <- R4+0
.loc 1 98 29 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:98:29
lsr.w r0, r8, #2
movs r1, #4
blx _calloc
mov r5, r0
Ltmp54:
@DEBUG_VALUE: outputBuffer <- R5+0
mov r0, r10
Ltmp55:
@DEBUG_VALUE: height <- R0+0
.loc 1 101 29 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:101:29
cmp r0, #0
Ltmp56:
@DEBUG_VALUE: rowIndex <- 0+0
beq LBB4_3
@ BB#1: @ %.lr.ph
Ltmp57:
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:self <- R0+0
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:_cmd <- R1+0
@DEBUG_VALUE: width <- [sp+#16]+#0
@DEBUG_VALUE: height <- R0+0
@DEBUG_VALUE: bytesPerRow <- [sp+#12]+#0
@DEBUG_VALUE: imageAlpha <- [sp+#4]+#0
@DEBUG_VALUE: data <- [sp+#8]+#0
@DEBUG_VALUE: buffer <- R4+0
@DEBUG_VALUE: outputBuffer <- R5+0
@DEBUG_VALUE: rowIndex <- 0+0
ldr r1, [sp, #12] @ 4-byte Reload
Ltmp58:
@DEBUG_VALUE: bytesPerRow <- R1+0
mov.w r8, #0
lsl.w r11, r1, #1
.loc 1 104 74 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:104:74
Ltmp59:
lsr.w r10, r1, #1
Ltmp60:
@DEBUG_VALUE: bytesPerRow <- [sp+#12]+#0
LBB4_2: @ =>This Inner Loop Header: Depth=1
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:self <- R0+0
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:_cmd <- R1+0
@DEBUG_VALUE: width <- [sp+#16]+#0
@DEBUG_VALUE: height <- R0+0
@DEBUG_VALUE: bytesPerRow <- [sp+#12]+#0
@DEBUG_VALUE: imageAlpha <- [sp+#4]+#0
@DEBUG_VALUE: data <- [sp+#8]+#0
@DEBUG_VALUE: outputBuffer <- R5+0
@DEBUG_VALUE: rowIndex <- 0+0
lsr.w r1, r8, #1
Ltmp61:
mov r6, r0
Ltmp62:
@DEBUG_VALUE: height <- R6+0
mla r0, r1, r10, r5
Ltmp63:
@DEBUG_VALUE: destRow <- R1+0
.loc 1 105 9 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:105:9
ldr r2, [sp, #16] @ 4-byte Reload
mov r1, r4
Ltmp64:
bl _resizeRow
mov r0, r6
Ltmp65:
@DEBUG_VALUE: height <- R0+0
.loc 1 101 50 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:101:50
add.w r8, r8, #2
Ltmp66:
@DEBUG_VALUE: rowIndex <- R8+0
.loc 1 101 29 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:101:29
add r4, r11
cmp r8, r0
blo LBB4_2
Ltmp67:
LBB4_3: @ %._crit_edge
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:self <- R0+0
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:_cmd <- R1+0
@DEBUG_VALUE: width <- [sp+#16]+#0
@DEBUG_VALUE: height <- R0+0
@DEBUG_VALUE: bytesPerRow <- [sp+#12]+#0
@DEBUG_VALUE: imageAlpha <- [sp+#4]+#0
@DEBUG_VALUE: data <- [sp+#8]+#0
@DEBUG_VALUE: outputBuffer <- R5+0
.loc 1 109 28 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:109:28
ldr r1, [sp, #4] @ 4-byte Reload
Ltmp68:
lsrs r2, r0, #1
str r1, [sp]
mov r6, r5
Ltmp69:
@DEBUG_VALUE: outputBuffer <- R6+0
ldr r1, [sp, #16] @ 4-byte Reload
ldr r0, [sp, #12] @ 4-byte Reload
Ltmp70:
lsrs r1, r1, #1
lsrs r3, r0, #1
mov r0, r5
bl _createBitmapContext
mov r4, r0
Ltmp71:
@DEBUG_VALUE: context <- R4+0
.loc 1 110 30 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:110:30
blx _CGBitmapContextCreateImage
.loc 1 111 66 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:111:66
movw r1, :lower16:(L_OBJC_SELECTOR_REFERENCES_4-(LPC4_1+4))
.loc 1 110 30 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:110:30
mov r5, r0
Ltmp72:
@DEBUG_VALUE: scaledImage <- R5+0
.loc 1 111 66 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:111:66
movt r1, :upper16:(L_OBJC_SELECTOR_REFERENCES_4-(LPC4_1+4))
movw r0, :lower16:(L_OBJC_CLASSLIST_REFERENCES_$_-(LPC4_2+4))
movt r0, :upper16:(L_OBJC_CLASSLIST_REFERENCES_$_-(LPC4_2+4))
LPC4_1:
add r1, pc
LPC4_2:
add r0, pc
mov r2, r5
ldr r1, [r1]
ldr r0, [r0]
blx _objc_msgSend
Ltmp73:
@DEBUG_VALUE: returnImage <- R0+0
@ InlineAsm Start
mov r7, r7 @ marker for objc_retainAutoreleaseReturnValue
@ InlineAsm End
blx _objc_retainAutoreleasedReturnValue
Ltmp74:
mov r8, r0
.loc 1 112 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:112:5
mov r0, r5
blx _CGImageRelease
.loc 1 113 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:113:5
mov r0, r4
blx _CGContextRelease
.loc 1 114 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:114:5
ldr r0, [sp, #8] @ 4-byte Reload
blx _CFRelease
.loc 1 115 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:115:5
mov r0, r6
blx _free
Ltmp75:
.loc 1 118 1 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:118:1
mov r0, r8
add sp, #20
pop.w {r8, r10, r11}
pop.w {r4, r5, r6, r7, lr}
Ltmp76:
b.w _objc_autoreleaseReturnValue
Ltmp77:
Lfunc_end4:
.cfi_endproc
.align 2
.code 16 @ @resizeRow
.thumb_func _resizeRow
_resizeRow:
.cfi_startproc
Lfunc_begin5:
.loc 1 26 0 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:26:0
@ BB#0:
@DEBUG_VALUE: resizeRow:dst <- R0+0
@DEBUG_VALUE: resizeRow:src <- R1+0
@DEBUG_VALUE: resizeRow:pixelsPerRow <- R2+0
.loc 1 27 47 prologue_end @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:27:47
add.w r3, r1, r2, lsl #2
Ltmp78:
@DEBUG_VALUE: rowB <- R3+0
.loc 1 30 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:30:5
bic r2, r2, #7
Ltmp79:
.loc 1 32 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:32:5
@ InlineAsm Start
Lresizeloop:
vld1.32 {d0-d3}, [r1]!
vld1.32 {d4-d7}, [r3]!
vhadd.u8 q0, q0, q2
vhadd.u8 q1, q1, q3
vtrn.32 q0, q2
vtrn.32 q1, q3
vhadd.u8 q0, q0, q2
vhadd.u8 q1, q1, q3
vtrn.32 d0, d1
vtrn.32 d2, d3
vswp d1, d2
vst1.64 {d0-d1}, [r0]!
subs r2, r2, #8
bne Lresizeloop
@ InlineAsm End
Ltmp80:
.loc 1 51 1 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:51:1
bx lr
Ltmp81:
Lfunc_end5:
.cfi_endproc
，O2的非功能输出如下：
.align 2
.code 16 @ @"\01-[BDPViewController downscaleImageNeon:]"
.thumb_func "-[BDPViewController downscaleImageNeon:]"
"-[BDPViewController downscaleImageNeon:]":
.cfi_startproc
Lfunc_begin4:
.loc 1 86 0 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:86:0
@ BB#0:
.loc 1 86 1 prologue_end @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:86:1
push {r4, r5, r6, r7, lr}
add r7, sp, #12
push.w {r8, r10, r11}
sub sp, #20
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:self <- R0+0
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:_cmd <- R1+0
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:image <- R2+0
.loc 1 88 20 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:88:20
Ltmp41:
movw r0, :lower16:(L_OBJC_SELECTOR_REFERENCES_2-(LPC4_0+4))
Ltmp42:
mov r6, r2
Ltmp43:
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:image <- R6+0
movt r0, :upper16:(L_OBJC_SELECTOR_REFERENCES_2-(LPC4_0+4))
LPC4_0:
add r0, pc
ldr.w r11, [r0]
mov r0, r6
blx _objc_retain
mov r4, r0
mov r0, r6
mov r1, r11
Ltmp44:
blx _objc_msgSend
blx _CGImageGetWidth
mov r5, r0
Ltmp45:
@DEBUG_VALUE: width <- R5+0
.loc 1 89 21 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:89:21
mov r0, r6
mov r1, r11
str r5, [sp, #16] @ 4-byte Spill
blx _objc_msgSend
blx _CGImageGetHeight
mov r10, r0
Ltmp46:
@DEBUG_VALUE: height <- R10+0
.loc 1 90 26 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:90:26
mov r0, r6
mov r1, r11
blx _objc_msgSend
blx _CGImageGetBytesPerRow
str r0, [sp, #12] @ 4-byte Spill
Ltmp47:
@DEBUG_VALUE: bytesPerRow <- [sp+#12]+#0
.loc 1 91 35 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:91:35
mov r0, r6
mov r1, r11
blx _objc_msgSend
blx _CGImageGetAlphaInfo
str r0, [sp, #4] @ 4-byte Spill
Ltmp48:
@DEBUG_VALUE: imageAlpha <- [sp+#4]+#0
.loc 1 94 45 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:94:45
mov r0, r6
mov r1, r11
blx _objc_msgSend
mov r6, r0
Ltmp49:
mov r0, r4
blx _objc_release
mov r0, r6
.loc 1 98 29 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:98:29
mul r8, r10, r5
Ltmp50:
@DEBUG_VALUE: width <- [sp+#16]+#0
.loc 1 94 45 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:94:45
blx _CGImageGetDataProvider
blx _CGDataProviderCopyData
Ltmp51:
@DEBUG_VALUE: data <- R0+0
str r0, [sp, #8] @ 4-byte Spill
Ltmp52:
@DEBUG_VALUE: data <- [sp+#8]+#0
.loc 1 95 29 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:95:29
blx _CFDataGetBytePtr
mov r4, r0
Ltmp53:
@DEBUG_VALUE: buffer <- R4+0
.loc 1 98 29 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:98:29
lsr.w r0, r8, #2
movs r1, #4
blx _calloc
mov r5, r0
Ltmp54:
@DEBUG_VALUE: outputBuffer <- R5+0
mov r0, r10
Ltmp55:
@DEBUG_VALUE: height <- R0+0
.loc 1 101 29 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:101:29
cmp r0, #0
Ltmp56:
@DEBUG_VALUE: rowIndex <- 0+0
beq LBB4_3
@ BB#1: @ %.lr.ph
Ltmp57:
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:self <- R0+0
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:_cmd <- R1+0
@DEBUG_VALUE: width <- [sp+#16]+#0
@DEBUG_VALUE: height <- R0+0
@DEBUG_VALUE: bytesPerRow <- [sp+#12]+#0
@DEBUG_VALUE: imageAlpha <- [sp+#4]+#0
@DEBUG_VALUE: data <- [sp+#8]+#0
@DEBUG_VALUE: buffer <- R4+0
@DEBUG_VALUE: outputBuffer <- R5+0
@DEBUG_VALUE: rowIndex <- 0+0
ldr r1, [sp, #12] @ 4-byte Reload
Ltmp58:
@DEBUG_VALUE: bytesPerRow <- R1+0
mov.w r8, #0
lsl.w r11, r1, #1
.loc 1 104 74 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:104:74
Ltmp59:
lsr.w r10, r1, #1
Ltmp60:
@DEBUG_VALUE: bytesPerRow <- [sp+#12]+#0
LBB4_2: @ =>This Inner Loop Header: Depth=1
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:self <- R0+0
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:_cmd <- R1+0
@DEBUG_VALUE: width <- [sp+#16]+#0
@DEBUG_VALUE: height <- R0+0
@DEBUG_VALUE: bytesPerRow <- [sp+#12]+#0
@DEBUG_VALUE: imageAlpha <- [sp+#4]+#0
@DEBUG_VALUE: data <- [sp+#8]+#0
@DEBUG_VALUE: outputBuffer <- R5+0
@DEBUG_VALUE: rowIndex <- 0+0
lsr.w r1, r8, #1
Ltmp61:
mov r6, r0
Ltmp62:
@DEBUG_VALUE: height <- R6+0
mla r0, r1, r10, r5
Ltmp63:
@DEBUG_VALUE: destRow <- R1+0
.loc 1 105 9 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:105:9
ldr r2, [sp, #16] @ 4-byte Reload
mov r1, r4
Ltmp64:
bl _resizeRow
mov r0, r6
Ltmp65:
@DEBUG_VALUE: height <- R0+0
.loc 1 101 50 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:101:50
add.w r8, r8, #2
Ltmp66:
@DEBUG_VALUE: rowIndex <- R8+0
.loc 1 101 29 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:101:29
add r4, r11
cmp r8, r0
blo LBB4_2
Ltmp67:
LBB4_3: @ %._crit_edge
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:self <- R0+0
@DEBUG_VALUE: -[BDPViewController downscaleImageNeon:]:_cmd <- R1+0
@DEBUG_VALUE: width <- [sp+#16]+#0
@DEBUG_VALUE: height <- R0+0
@DEBUG_VALUE: bytesPerRow <- [sp+#12]+#0
@DEBUG_VALUE: imageAlpha <- [sp+#4]+#0
@DEBUG_VALUE: data <- [sp+#8]+#0
@DEBUG_VALUE: outputBuffer <- R5+0
.loc 1 109 28 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:109:28
ldr r1, [sp, #4] @ 4-byte Reload
Ltmp68:
lsrs r2, r0, #1
str r1, [sp]
mov r6, r5
Ltmp69:
@DEBUG_VALUE: outputBuffer <- R6+0
ldr r1, [sp, #16] @ 4-byte Reload
ldr r0, [sp, #12] @ 4-byte Reload
Ltmp70:
lsrs r1, r1, #1
lsrs r3, r0, #1
mov r0, r5
bl _createBitmapContext
mov r4, r0
Ltmp71:
@DEBUG_VALUE: context <- R4+0
.loc 1 110 30 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:110:30
blx _CGBitmapContextCreateImage
.loc 1 111 66 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:111:66
movw r1, :lower16:(L_OBJC_SELECTOR_REFERENCES_4-(LPC4_1+4))
.loc 1 110 30 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:110:30
mov r5, r0
Ltmp72:
@DEBUG_VALUE: scaledImage <- R5+0
.loc 1 111 66 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:111:66
movt r1, :upper16:(L_OBJC_SELECTOR_REFERENCES_4-(LPC4_1+4))
movw r0, :lower16:(L_OBJC_CLASSLIST_REFERENCES_$_-(LPC4_2+4))
movt r0, :upper16:(L_OBJC_CLASSLIST_REFERENCES_$_-(LPC4_2+4))
LPC4_1:
add r1, pc
LPC4_2:
add r0, pc
mov r2, r5
ldr r1, [r1]
ldr r0, [r0]
blx _objc_msgSend
Ltmp73:
@DEBUG_VALUE: returnImage <- R0+0
@ InlineAsm Start
mov r7, r7 @ marker for objc_retainAutoreleaseReturnValue
@ InlineAsm End
blx _objc_retainAutoreleasedReturnValue
Ltmp74:
mov r8, r0
.loc 1 112 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:112:5
mov r0, r5
blx _CGImageRelease
.loc 1 113 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:113:5
mov r0, r4
blx _CGContextRelease
.loc 1 114 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:114:5
ldr r0, [sp, #8] @ 4-byte Reload
blx _CFRelease
.loc 1 115 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:115:5
mov r0, r6
blx _free
Ltmp75:
.loc 1 118 1 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:118:1
mov r0, r8
add sp, #20
pop.w {r8, r10, r11}
pop.w {r4, r5, r6, r7, lr}
Ltmp76:
b.w _objc_autoreleaseReturnValue
Ltmp77:
Lfunc_end4:
.cfi_endproc
.align 2
.code 16 @ @resizeRow
.thumb_func _resizeRow
_resizeRow:
.cfi_startproc
Lfunc_begin5:
.loc 1 26 0 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:26:0
@ BB#0:
@DEBUG_VALUE: resizeRow:dst <- R0+0
@DEBUG_VALUE: resizeRow:src <- R1+0
@DEBUG_VALUE: resizeRow:pixelsPerRow <- R2+0
.loc 1 27 47 prologue_end @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:27:47
add.w r3, r1, r2, lsl #2
Ltmp78:
@DEBUG_VALUE: rowB <- R3+0
.loc 1 30 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:30:5
bic r2, r2, #7
Ltmp79:
.loc 1 32 5 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:32:5
@ InlineAsm Start
Lresizeloop:
vld1.32 {d0-d3}, [r1]!
vld1.32 {d4-d7}, [r3]!
vhadd.u8 q0, q0, q2
vhadd.u8 q1, q1, q3
vtrn.32 q0, q2
vtrn.32 q1, q3
vhadd.u8 q0, q0, q2
vhadd.u8 q1, q1, q3
vtrn.32 d0, d1
vtrn.32 d2, d3
vswp d1, d2
vst1.64 {d0-d1}, [r0]!
subs r2, r2, #8
bne Lresizeloop
@ InlineAsm End
Ltmp80:
.loc 1 51 1 @ NEON-Image-Downscaling/ImageResize/BDPViewController.m:51:1
bx lr
Ltmp81:
Lfunc_end5:
.cfi_endproc
答案 0 :(得分：13)
以下是我使用-O2从Xcode项目获得的汇编代码片段。 （使用-O1构建并不打算内联函数，所以我并不感到惊讶它工作正常。）
Ltmp55:
@DEBUG_VALUE: rowIndex <- R3+0
.loc 1 101 29 @ /tmp/NEON-Image-Downscaling/ImageResize/BDPViewController.m:101:29
add r8, r12
cmp r3, r11
.loc 1 32 5 @ /tmp/NEON-Image-Downscaling/ImageResize/BDPViewController.m:32:5
Ltmp56:
@ InlineAsm Start
Lresizeloop:
vld1.32 {d0-d3}, [r4]!
vld1.32 {d4-d7}, [r5]!
vhadd.u8 q0, q0, q2
vhadd.u8 q1, q1, q3
vtrn.32 q0, q2
vtrn.32 q1, q3
vhadd.u8 q0, q0, q2
vhadd.u8 q1, q1, q3
vtrn.32 d0, d1
vtrn.32 d2, d3
vswp d1, d2
vst1.64 {d0-d1}, [r6]!
subs r2, r2, #8
bne Lresizeloop
@ InlineAsm End
Ltmp57:
blo LBB2_2
在最后一行看到blo（branch-if-lower）指令？它使用由程序集块顶部的cmp r3, r11设置的条件代码。但是当然你的内联汇编代码完全破坏了条件代码寄存器。那么这是一个编译器错误吗？... 没有！你只是忘了告诉编译器你的内联汇编代码会破坏条件代码。取代
: "=r"(dst), "=r"(src), "=r"(rowB), "=r"(pixelsPerRow)
: "0"(dst), "1"(src), "2"(rowB), "3"(pixelsPerRow)
: "q0", "q1", "q2", "q3"
);
与
: "=r"(dst), "=r"(src), "=r"(rowB), "=r"(pixelsPerRow)
: "0"(dst), "1"(src), "2"(rowB), "3"(pixelsPerRow)
: "q0", "q1", "q2", "q3", "cc"
);
并且程序集输出自行修复。我没有运行应用程序，但我打赌你会发现它现在好多了。 ：）