On Wednesday, 28 November 2012 at 00:19:33 UTC, bearophile wrote:
dpaste is currently down and I don't have gdc2/dc2 installed
here, so I can't show the asm from those other compilers.
fwiw, here is what LDC produces:
---
_D4test18__T10reverseArrTiZ10reverseArrFAiZv:
lea RAX, QWORD PTR [RSI + 4*RDI - 4]
cmp RSI, RAX
jae .LBB1_3
lea RAX, QWORD PTR [RSI + 4*RDI - 8]
.align 16, 0x90
.LBB1_2:
mov ECX, DWORD PTR [RSI]
mov EDX, DWORD PTR [RAX + 4]
mov DWORD PTR [RSI], EDX
mov DWORD PTR [RAX + 4], ECX
add RSI, 4
cmp RSI, RAX
lea RAX, QWORD PTR [RAX - 4]
jb .LBB1_2
.LBB1_3:
ret
---
---
_D3std9algorithm15__T7reverseTAiZ7reverseFAiZv:
test RDI, RDI
je .LBB2_4
lea RAX, QWORD PTR [RSI + 4*RDI - 4]
.align 16, 0x90
.LBB2_2:
mov ECX, DWORD PTR [RSI]
mov EDX, DWORD PTR [RAX]
mov DWORD PTR [RSI], EDX
mov DWORD PTR [RAX], ECX
cmp RDI, 1
je .LBB2_4
add RAX, -4
add RSI, 4
add RDI, -2
jne .LBB2_2
.LBB2_4:
ret
---
---
_D4test20__T12reverseArrayTiZ12reverseArrayFAiZv:
lea RAX, QWORD PTR [RSI + 4*RDI - 4]
cmp RSI, RAX
jae .LBB3_3
add RSI, 4
.align 16, 0x90
.LBB3_2:
mov ECX, DWORD PTR [RSI - 4]
mov EDX, DWORD PTR [RAX]
mov DWORD PTR [RSI - 4], EDX
mov DWORD PTR [RAX], ECX
add RAX, -4
cmp RSI, RAX
lea RSI, QWORD PTR [RSI + 4]
jb .LBB3_2
.LBB3_3:
ret
---
The extra jump in the std.algorithm version is still there – I
haven't checked whether it would be feasible to optimize the
induction variable away entirely.
David
