Posted to tcl by aspect at Wed Sep 30 00:10:41 GMT 2015view pretty
Examining the assembly from applying drh's example optimisation to CompareVarKeys.
Assembly obtained with:
configure --enable-symbols=all
make
rm tclVar.o
make -n tclVar.o | sed -e 's/^gcc/gcc -S -fverbose-asm/' | sh
Original 7b3b0ca73eb7ba1cbce9 (without optimisation):
.type CompareVarKeys, @function
CompareVarKeys:
.LFB82:
.loc 1 6359 0
.cfi_startproc
pushq %rbp #
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movq %rsp, %rbp #,
.cfi_def_cfa_register 6
pushq %r14 #
pushq %r13 #
pushq %r12 #
pushq %rbx #
subq $32, %rsp #,
.cfi_offset 14, -24
.cfi_offset 13, -32
.cfi_offset 12, -40
.cfi_offset 3, -48
movq %rdi, -56(%rbp) # keyPtr, keyPtr
movq %rsi, -64(%rbp) # hPtr, hPtr
.loc 1 6360 0
movq -56(%rbp), %rax # keyPtr, tmp95
movq %rax, -40(%rbp) # tmp95, objPtr1
.loc 1 6361 0
movq -64(%rbp), %rax # hPtr, tmp96
movq 32(%rax), %rax # hPtr_7(D)->key.objPtr, tmp97
movq %rax, -48(%rbp) # tmp97, objPtr2
.loc 1 6369 0
movq -40(%rbp), %rax # objPtr1, tmp98
cmpq -48(%rbp), %rax # objPtr2, tmp98
jne .L910 #,
.loc 1 6370 0
movl $1, %eax #, D.20671
jmp .L911 #
.L910:
.loc 1 6378 0
movq -40(%rbp), %rax # objPtr1, tmp99
movq 8(%rax), %rax # objPtr1_6->bytes, D.20670
testq %rax, %rax # D.20670
je .L912 #,
.loc 1 6378 0 is_stmt 0 discriminator 1
movq -40(%rbp), %rax # objPtr1, tmp100
movq 8(%rax), %rax # objPtr1_6->bytes, D.20670
jmp .L913 #
.L912:
.loc 1 6378 0 discriminator 2
movq -40(%rbp), %rax # objPtr1, tmp101
movq %rax, %rdi # tmp101,
call Tcl_GetString@PLT #
.L913:
.loc 1 6378 0 discriminator 4
movq %rax, %r13 # D.20670, p1
.loc 1 6379 0 is_stmt 1 discriminator 4
movq -40(%rbp), %rax # objPtr1, tmp102
movl 16(%rax), %ebx # objPtr1_6->length, l1
.loc 1 6380 0 discriminator 4
movq -48(%rbp), %rax # objPtr2, tmp103
movq 8(%rax), %rax # objPtr2_8->bytes, D.20670
testq %rax, %rax # D.20670
je .L914 #,
.loc 1 6380 0 is_stmt 0 discriminator 1
movq -48(%rbp), %rax # objPtr2, tmp104
movq 8(%rax), %rax # objPtr2_8->bytes, D.20670
jmp .L915 #
.L914:
.loc 1 6380 0 discriminator 2
movq -48(%rbp), %rax # objPtr2, tmp105
movq %rax, %rdi # tmp105,
call Tcl_GetString@PLT #
.L915:
.loc 1 6380 0 discriminator 4
movq %rax, %r14 # D.20670, p2
.loc 1 6381 0 is_stmt 1 discriminator 4
movq -48(%rbp), %rax # objPtr2, tmp106
movl 16(%rax), %r12d # objPtr2_8->length, l2
.loc 1 6387 0 discriminator 4
cmpl %r12d, %ebx # l2, l1
jne .L916 #,
.loc 1 6387 0 is_stmt 0 discriminator 1
movslq %ebx, %rax # l1, D.20672
movq %rax, %rdx # D.20672,
movq %r14, %rsi # p2,
movq %r13, %rdi # p1,
call memcmp@PLT #
testl %eax, %eax # D.20671
jne .L916 #,
.loc 1 6387 0 discriminator 3
movl $1, %eax #, D.20671
jmp .L918 #
.L916:
.loc 1 6387 0 discriminator 4
movl $0, %eax #, D.20671
.L918:
.loc 1 6387 0
nop
.L911:
.loc 1 6388 0 is_stmt 1
addq $32, %rsp #,
popq %rbx #
popq %r12 #
popq %r13 #
popq %r14 #
popq %rbp #
.cfi_def_cfa 7, 8
ret
.cfi_endproc
.LFE82:
.size CompareVarKeys, .-CompareVarKeys
Optimised outer function:
.type CompareVarKeys, @function
CompareVarKeys:
.LFB83:
.loc 1 6392 0
.cfi_startproc
pushq %rbp #
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movq %rsp, %rbp #,
.cfi_def_cfa_register 6
subq $32, %rsp #,
movq %rdi, -24(%rbp) # keyPtr, keyPtr
movq %rsi, -32(%rbp) # hPtr, hPtr
.loc 1 6393 0
movq -24(%rbp), %rax # keyPtr, tmp85
movq %rax, -8(%rbp) # tmp85, objPtr1
.loc 1 6394 0
movq -32(%rbp), %rax # hPtr, tmp86
movq 32(%rax), %rax # hPtr_4(D)->key.objPtr, tmp87
movq %rax, -16(%rbp) # tmp87, objPtr2
.loc 1 6400 0
movq -8(%rbp), %rax # objPtr1, tmp88
cmpq -16(%rbp), %rax # objPtr2, tmp88
jne .L918 #,
.loc 1 6401 0
movl $1, %eax #, D.20680
jmp .L919 #
.L918:
.loc 1 6403 0
movq -16(%rbp), %rdx # objPtr2, tmp89
movq -8(%rbp), %rax # objPtr1, tmp90
movq %rdx, %rsi # tmp89,
movq %rax, %rdi # tmp90,
call CompareDistinctVarKeys #
.L919:
.loc 1 6405 0
leave
.cfi_def_cfa 7, 8
ret
.cfi_endproc
.LFE83:
.size CompareVarKeys, .-CompareVarKeys
Inner function:
.type CompareDistinctVarKeys, @function
CompareDistinctVarKeys:
.LFB82:
.loc 1 6367 0
.cfi_startproc
pushq %rbp #
.cfi_def_cfa_offset 16
.cfi_offset 6, -16
movq %rsp, %rbp #,
.cfi_def_cfa_register 6
pushq %r14 #
pushq %r13 #
pushq %r12 #
pushq %rbx #
subq $16, %rsp #,
.cfi_offset 14, -24
.cfi_offset 13, -32
.cfi_offset 12, -40
.cfi_offset 3, -48
movq %rdi, -40(%rbp) # objPtr1, objPtr1
movq %rsi, -48(%rbp) # objPtr2, objPtr2
.loc 1 6376 0
movq -40(%rbp), %rax # objPtr1, tmp95
movq 8(%rax), %rax # objPtr1_4(D)->bytes, D.20677
testq %rax, %rax # D.20677
je .L910 #,
.loc 1 6376 0 is_stmt 0 discriminator 1
movq -40(%rbp), %rax # objPtr1, tmp96
movq 8(%rax), %rax # objPtr1_4(D)->bytes, D.20677
jmp .L911 #
.L910:
.loc 1 6376 0 discriminator 2
movq -40(%rbp), %rax # objPtr1, tmp97
movq %rax, %rdi # tmp97,
call Tcl_GetString@PLT #
.L911:
.loc 1 6376 0 discriminator 4
movq %rax, %r13 # D.20677, p1
.loc 1 6377 0 is_stmt 1 discriminator 4
movq -40(%rbp), %rax # objPtr1, tmp98
movl 16(%rax), %ebx # objPtr1_4(D)->length, l1
.loc 1 6378 0 discriminator 4
movq -48(%rbp), %rax # objPtr2, tmp99
movq 8(%rax), %rax # objPtr2_10(D)->bytes, D.20677
testq %rax, %rax # D.20677
je .L912 #,
.loc 1 6378 0 is_stmt 0 discriminator 1
movq -48(%rbp), %rax # objPtr2, tmp100
movq 8(%rax), %rax # objPtr2_10(D)->bytes, D.20677
jmp .L913 #
.L912:
.loc 1 6378 0 discriminator 2
movq -48(%rbp), %rax # objPtr2, tmp101
movq %rax, %rdi # tmp101,
call Tcl_GetString@PLT #
.L913:
.loc 1 6378 0 discriminator 4
movq %rax, %r14 # D.20677, p2
.loc 1 6379 0 is_stmt 1 discriminator 4
movq -48(%rbp), %rax # objPtr2, tmp102
movl 16(%rax), %r12d # objPtr2_10(D)->length, l2
.loc 1 6385 0 discriminator 4
cmpl %r12d, %ebx # l2, l1
jne .L914 #,
.loc 1 6385 0 is_stmt 0 discriminator 1
movslq %ebx, %rax # l1, D.20679
movq %rax, %rdx # D.20679,
movq %r14, %rsi # p2,
movq %r13, %rdi # p1,
call memcmp@PLT #
testl %eax, %eax # D.20678
jne .L914 #,
.loc 1 6385 0 discriminator 3
movl $1, %eax #, D.20678
jmp .L915 #
.L914:
.loc 1 6385 0 discriminator 4
movl $0, %eax #, D.20678
.L915:
.loc 1 6386 0 is_stmt 1 discriminator 6
addq $16, %rsp #,
popq %rbx #
popq %r12 #
popq %r13 #
popq %r14 #
popq %rbp #
.cfi_def_cfa 7, 8
ret
.cfi_endproc
.LFE82:
.size CompareDistinctVarKeys, .-CompareDistinctVarKeys