Compile the attached source code with options -Os -fpic -mthumb, gcc generates
following code for the constructor:
push {r4, r5, r6, r7, lr}
.LCFI2:
.pad #12
sub sp, sp, #12
.LCFI3:
str r3, [sp]
add r3, sp, #32 //A, B
ldrb r3, [r3] //A, B
ldr r5, .L14
mov r4, r0
mov r7, r2
mov r6, r1
str r3, [sp, #4] //A
bl _ZN1AC2Ev
ldr r3, .L14+4
.LPIC2:
add r5, pc
ldr r3, [r5, r3]
add r3, r3, #8
mov r1, sp //A
str r3, [r4]
add r2, r1, #4 //A, C
mov r3, r4
add r3, r3, #42
ldrb r1, [r2] //A, C
strb r7, [r3]
add r3, r3, #1 // C
strb r1, [r3] //A, C
cmp r6, #0
beq .L9
ldr r2, [sp]
mov r3, #2
cmp r2, #0
bne .L12
mov r3, #1
.L12:
strh r3, [r4, #40]
b .L11
.L9:
strh r6, [r4, #40]
.L11:
add sp, sp, #12
mov r0, r4
@ sp needed for prologue
pop {r4, r5, r6, r7}
pop {r1}
bx r1
1. The instructions marked A move a byte from parameter to the object body, but
it first moves that parameter to stack [sp, 4], then move the value from [sp,4]
to the target object. The move to [sp, 4] is redundant.
2. The Instructions marked B load a byte from the stack, then store the whole
word into another place on the stack. Since later instructions load a byte from
the new space, we can simply move the whole word at the first place, so
instructions marked as B can be reduced to one instruction "ldr r3, [sp, 32]"
3. Instructions marked C can be simplified to
ldrb r1, [r1, 4]
strb r1, [r3, 1]
4. Why not use "pop {r4, r5, r6, r7, pc}" as the epilogue?
--
Summary: redundant memory move from parameter space to spill
space
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: carrot at google dot com
GCC build triplet: i686-linux
GCC host triplet: i686-linux
GCC target triplet: arm-eabi
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42235
