maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/X86/lea-opt.ll

310 lines
11 KiB
LLVM
Raw Normal View History

[X86] Autogenerate complete checks and fix a failure introduced in r337875. llvm-svn: 337889
2018-07-25 13:22:13 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
[X86] Enable RRL part of the LEA optimization pass for -O2. Enable "Remove Redundant LEAs" part of the LEA optimization pass for -O2. This gives 6.4% performance improve on Broadwell on nnet benchmark from Coremark-pro. There is no significant effect on other benchmarks (Geekbench, Spec2000, Spec2006). Differential Revision: http://reviews.llvm.org/D19659 llvm-svn: 270036
2016-05-19 18:18:29 +08:00
; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s -check-prefix=CHECK -check-prefix=ENABLED
; RUN: llc --disable-x86-lea-opt < %s -mtriple=x86_64-linux | FileCheck %s -check-prefix=CHECK -check-prefix=DISABLED
LEA code size optimization pass (Part 1): Remove redundant address recalculations, by Andrey Turetsky Add new x86 pass which replaces address calculations in load or store instructions with def register of existing LEA (must be in the same basic block), if the LEA calculates address that differs only by a displacement. Works only with -Os or -Oz. Differential Revision: http://reviews.llvm.org/D13294 llvm-svn: 254712
2015-12-04 18:53:15 +08:00
%struct.anon1 = type { i32, i32, i32 }
%struct.anon2 = type { i32, [32 x i32], i32 }
@arr1 = external global [65 x %struct.anon1], align 16
@arr2 = external global [65 x %struct.anon2], align 16
define void @test1(i64 %x) nounwind {
[X86] Autogenerate complete checks and fix a failure introduced in r337875. llvm-svn: 337889
2018-07-25 13:22:13 +08:00
; ENABLED-LABEL: test1:
; ENABLED: # %bb.0: # %entry
; ENABLED-NEXT: shlq $2, %rdi
; ENABLED-NEXT: movl arr1(%rdi,%rdi,2), %ecx
; ENABLED-NEXT: leaq arr1+4(%rdi,%rdi,2), %rax
; ENABLED-NEXT: subl arr1+4(%rdi,%rdi,2), %ecx
; ENABLED-NEXT: addl arr1+8(%rdi,%rdi,2), %ecx
; ENABLED-NEXT: cmpl $2, %ecx
; ENABLED-NEXT: je .LBB0_3
; ENABLED-NEXT: # %bb.1: # %entry
; ENABLED-NEXT: cmpl $1, %ecx
; ENABLED-NEXT: jne .LBB0_4
; ENABLED-NEXT: # %bb.2: # %sw.bb.1
; ENABLED-NEXT: movl $111, (%rax)
; ENABLED-NEXT: movl $222, 4(%rax)
; ENABLED-NEXT: retq
; ENABLED-NEXT: .LBB0_3: # %sw.bb.2
; ENABLED-NEXT: movl $333, (%rax) # imm = 0x14D
; ENABLED-NEXT: movl $444, 4(%rax) # imm = 0x1BC
; ENABLED-NEXT: .LBB0_4: # %sw.epilog
; ENABLED-NEXT: retq
;
; DISABLED-LABEL: test1:
; DISABLED: # %bb.0: # %entry
; DISABLED-NEXT: shlq $2, %rdi
; DISABLED-NEXT: movl arr1(%rdi,%rdi,2), %edx
; DISABLED-NEXT: leaq arr1+4(%rdi,%rdi,2), %rax
; DISABLED-NEXT: subl arr1+4(%rdi,%rdi,2), %edx
; DISABLED-NEXT: leaq arr1+8(%rdi,%rdi,2), %rcx
; DISABLED-NEXT: addl arr1+8(%rdi,%rdi,2), %edx
; DISABLED-NEXT: cmpl $2, %edx
; DISABLED-NEXT: je .LBB0_3
; DISABLED-NEXT: # %bb.1: # %entry
; DISABLED-NEXT: cmpl $1, %edx
; DISABLED-NEXT: jne .LBB0_4
; DISABLED-NEXT: # %bb.2: # %sw.bb.1
; DISABLED-NEXT: movl $111, (%rax)
; DISABLED-NEXT: movl $222, (%rcx)
; DISABLED-NEXT: retq
; DISABLED-NEXT: .LBB0_3: # %sw.bb.2
; DISABLED-NEXT: movl $333, (%rax) # imm = 0x14D
; DISABLED-NEXT: movl $444, (%rcx) # imm = 0x1BC
; DISABLED-NEXT: .LBB0_4: # %sw.epilog
; DISABLED-NEXT: retq
LEA code size optimization pass (Part 1): Remove redundant address recalculations, by Andrey Turetsky Add new x86 pass which replaces address calculations in load or store instructions with def register of existing LEA (must be in the same basic block), if the LEA calculates address that differs only by a displacement. Works only with -Os or -Oz. Differential Revision: http://reviews.llvm.org/D13294 llvm-svn: 254712
2015-12-04 18:53:15 +08:00
entry:
%a = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 0
%tmp = load i32, i32* %a, align 4
%b = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 1
%tmp1 = load i32, i32* %b, align 4
%sub = sub i32 %tmp, %tmp1
%c = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 2
%tmp2 = load i32, i32* %c, align 4
%add = add nsw i32 %sub, %tmp2
switch i32 %add, label %sw.epilog [
i32 1, label %sw.bb.1
i32 2, label %sw.bb.2
]
sw.bb.1: ; preds = %entry
store i32 111, i32* %b, align 4
store i32 222, i32* %c, align 4
br label %sw.epilog
sw.bb.2: ; preds = %entry
store i32 333, i32* %b, align 4
store i32 444, i32* %c, align 4
br label %sw.epilog
sw.epilog: ; preds = %sw.bb.2, %sw.bb.1, %entry
ret void
}
define void @test2(i64 %x) nounwind optsize {
[X86] Autogenerate complete checks and fix a failure introduced in r337875. llvm-svn: 337889
2018-07-25 13:22:13 +08:00
; ENABLED-LABEL: test2:
; ENABLED: # %bb.0: # %entry
; ENABLED-NEXT: shlq $2, %rdi
; ENABLED-NEXT: leaq arr1+4(%rdi,%rdi,2), %rax
; ENABLED-NEXT: movl -4(%rax), %ecx
; ENABLED-NEXT: subl (%rax), %ecx
; ENABLED-NEXT: addl 4(%rax), %ecx
; ENABLED-NEXT: cmpl $2, %ecx
; ENABLED-NEXT: je .LBB1_3
; ENABLED-NEXT: # %bb.1: # %entry
; ENABLED-NEXT: cmpl $1, %ecx
; ENABLED-NEXT: jne .LBB1_4
; ENABLED-NEXT: # %bb.2: # %sw.bb.1
; ENABLED-NEXT: movl $111, (%rax)
; ENABLED-NEXT: movl $222, 4(%rax)
; ENABLED-NEXT: retq
; ENABLED-NEXT: .LBB1_3: # %sw.bb.2
; ENABLED-NEXT: movl $333, (%rax) # imm = 0x14D
; ENABLED-NEXT: movl $444, 4(%rax) # imm = 0x1BC
; ENABLED-NEXT: .LBB1_4: # %sw.epilog
; ENABLED-NEXT: retq
;
; DISABLED-LABEL: test2:
; DISABLED: # %bb.0: # %entry
; DISABLED-NEXT: shlq $2, %rdi
; DISABLED-NEXT: movl arr1(%rdi,%rdi,2), %edx
; DISABLED-NEXT: leaq arr1+4(%rdi,%rdi,2), %rax
; DISABLED-NEXT: subl arr1+4(%rdi,%rdi,2), %edx
; DISABLED-NEXT: leaq arr1+8(%rdi,%rdi,2), %rcx
; DISABLED-NEXT: addl arr1+8(%rdi,%rdi,2), %edx
; DISABLED-NEXT: cmpl $2, %edx
; DISABLED-NEXT: je .LBB1_3
; DISABLED-NEXT: # %bb.1: # %entry
; DISABLED-NEXT: cmpl $1, %edx
; DISABLED-NEXT: jne .LBB1_4
; DISABLED-NEXT: # %bb.2: # %sw.bb.1
; DISABLED-NEXT: movl $111, (%rax)
; DISABLED-NEXT: movl $222, (%rcx)
; DISABLED-NEXT: retq
; DISABLED-NEXT: .LBB1_3: # %sw.bb.2
; DISABLED-NEXT: movl $333, (%rax) # imm = 0x14D
; DISABLED-NEXT: movl $444, (%rcx) # imm = 0x1BC
; DISABLED-NEXT: .LBB1_4: # %sw.epilog
; DISABLED-NEXT: retq
LEA code size optimization pass (Part 1): Remove redundant address recalculations, by Andrey Turetsky Add new x86 pass which replaces address calculations in load or store instructions with def register of existing LEA (must be in the same basic block), if the LEA calculates address that differs only by a displacement. Works only with -Os or -Oz. Differential Revision: http://reviews.llvm.org/D13294 llvm-svn: 254712
2015-12-04 18:53:15 +08:00
entry:
%a = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 0
%tmp = load i32, i32* %a, align 4
%b = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 1
%tmp1 = load i32, i32* %b, align 4
%sub = sub i32 %tmp, %tmp1
%c = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 2
%tmp2 = load i32, i32* %c, align 4
%add = add nsw i32 %sub, %tmp2
switch i32 %add, label %sw.epilog [
i32 1, label %sw.bb.1
i32 2, label %sw.bb.2
]
sw.bb.1: ; preds = %entry
store i32 111, i32* %b, align 4
store i32 222, i32* %c, align 4
br label %sw.epilog
sw.bb.2: ; preds = %entry
store i32 333, i32* %b, align 4
store i32 444, i32* %c, align 4
br label %sw.epilog
sw.epilog: ; preds = %sw.bb.2, %sw.bb.1, %entry
ret void
}
; Check that LEA optimization pass takes into account a resultant address
; displacement when choosing a LEA instruction for replacing a redundant
; address recalculation.
define void @test3(i64 %x) nounwind optsize {
[X86] Autogenerate complete checks and fix a failure introduced in r337875. llvm-svn: 337889
2018-07-25 13:22:13 +08:00
; ENABLED-LABEL: test3:
; ENABLED: # %bb.0: # %entry
; ENABLED-NEXT: movq %rdi, %rax
; ENABLED-NEXT: shlq $7, %rax
; ENABLED-NEXT: leaq arr2+132(%rax,%rdi,8), %rcx
; ENABLED-NEXT: leaq arr2(%rax,%rdi,8), %rax
; ENABLED-NEXT: movl (%rcx), %edx
; ENABLED-NEXT: addl (%rax), %edx
; ENABLED-NEXT: cmpl $2, %edx
; ENABLED-NEXT: je .LBB2_3
; ENABLED-NEXT: # %bb.1: # %entry
; ENABLED-NEXT: cmpl $1, %edx
; ENABLED-NEXT: jne .LBB2_4
; ENABLED-NEXT: # %bb.2: # %sw.bb.1
; ENABLED-NEXT: movl $111, (%rcx)
; ENABLED-NEXT: movl $222, (%rax)
; ENABLED-NEXT: retq
; ENABLED-NEXT: .LBB2_3: # %sw.bb.2
; ENABLED-NEXT: movl $333, (%rcx) # imm = 0x14D
; ENABLED-NEXT: movl %eax, (%rax)
; ENABLED-NEXT: .LBB2_4: # %sw.epilog
; ENABLED-NEXT: retq
;
; DISABLED-LABEL: test3:
; DISABLED: # %bb.0: # %entry
; DISABLED-NEXT: movq %rdi, %rsi
; DISABLED-NEXT: shlq $7, %rsi
; DISABLED-NEXT: leaq arr2+132(%rsi,%rdi,8), %rcx
; DISABLED-NEXT: leaq arr2(%rsi,%rdi,8), %rax
; DISABLED-NEXT: movl arr2+132(%rsi,%rdi,8), %edx
; DISABLED-NEXT: addl arr2(%rsi,%rdi,8), %edx
; DISABLED-NEXT: cmpl $2, %edx
; DISABLED-NEXT: je .LBB2_3
; DISABLED-NEXT: # %bb.1: # %entry
; DISABLED-NEXT: cmpl $1, %edx
; DISABLED-NEXT: jne .LBB2_4
; DISABLED-NEXT: # %bb.2: # %sw.bb.1
; DISABLED-NEXT: movl $111, (%rcx)
; DISABLED-NEXT: movl $222, (%rax)
; DISABLED-NEXT: retq
; DISABLED-NEXT: .LBB2_3: # %sw.bb.2
; DISABLED-NEXT: movl $333, (%rcx) # imm = 0x14D
; DISABLED-NEXT: movl %eax, (%rax)
; DISABLED-NEXT: .LBB2_4: # %sw.epilog
; DISABLED-NEXT: retq
LEA code size optimization pass (Part 1): Remove redundant address recalculations, by Andrey Turetsky Add new x86 pass which replaces address calculations in load or store instructions with def register of existing LEA (must be in the same basic block), if the LEA calculates address that differs only by a displacement. Works only with -Os or -Oz. Differential Revision: http://reviews.llvm.org/D13294 llvm-svn: 254712
2015-12-04 18:53:15 +08:00
entry:
%a = getelementptr inbounds [65 x %struct.anon2], [65 x %struct.anon2]* @arr2, i64 0, i64 %x, i32 2
%tmp = load i32, i32* %a, align 4
%b = getelementptr inbounds [65 x %struct.anon2], [65 x %struct.anon2]* @arr2, i64 0, i64 %x, i32 0
%tmp1 = load i32, i32* %b, align 4
%add = add nsw i32 %tmp, %tmp1
switch i32 %add, label %sw.epilog [
i32 1, label %sw.bb.1
i32 2, label %sw.bb.2
]
sw.bb.1: ; preds = %entry
store i32 111, i32* %a, align 4
store i32 222, i32* %b, align 4
br label %sw.epilog
sw.bb.2: ; preds = %entry
store i32 333, i32* %a, align 4
[X86] Enable RRL part of the LEA optimization pass for -O2. Enable "Remove Redundant LEAs" part of the LEA optimization pass for -O2. This gives 6.4% performance improve on Broadwell on nnet benchmark from Coremark-pro. There is no significant effect on other benchmarks (Geekbench, Spec2000, Spec2006). Differential Revision: http://reviews.llvm.org/D19659 llvm-svn: 270036
2016-05-19 18:18:29 +08:00
; Make sure the REG3's definition LEA won't be removed as redundant.
%cvt = ptrtoint i32* %b to i32
store i32 %cvt, i32* %b, align 4
LEA code size optimization pass (Part 1): Remove redundant address recalculations, by Andrey Turetsky Add new x86 pass which replaces address calculations in load or store instructions with def register of existing LEA (must be in the same basic block), if the LEA calculates address that differs only by a displacement. Works only with -Os or -Oz. Differential Revision: http://reviews.llvm.org/D13294 llvm-svn: 254712
2015-12-04 18:53:15 +08:00
br label %sw.epilog
sw.epilog: ; preds = %sw.bb.2, %sw.bb.1, %entry
ret void
; REG3's definition is closer to movl than REG2's, but the pass still chooses
; REG2 because it provides the resultant address displacement fitting 1 byte.
}
LEA code size optimization pass (Part 2): Remove redundant LEA instructions. Make x86 OptimizeLEAs pass remove LEA instruction if there is another LEA (in the same basic block) which calculates address differing only be a displacement. Works only for -Oz. Differential Revision: http://reviews.llvm.org/D13295 llvm-svn: 257589
2016-01-13 19:30:44 +08:00
define void @test4(i64 %x) nounwind minsize {
[X86] Autogenerate complete checks and fix a failure introduced in r337875. llvm-svn: 337889
2018-07-25 13:22:13 +08:00
; ENABLED-LABEL: test4:
; ENABLED: # %bb.0: # %entry
; ENABLED-NEXT: imulq $12, %rdi, %rax
; ENABLED-NEXT: leaq arr1+4(%rax), %rax
; ENABLED-NEXT: movl -4(%rax), %ecx
; ENABLED-NEXT: subl (%rax), %ecx
; ENABLED-NEXT: addl 4(%rax), %ecx
; ENABLED-NEXT: cmpl $2, %ecx
; ENABLED-NEXT: je .LBB3_3
; ENABLED-NEXT: # %bb.1: # %entry
; ENABLED-NEXT: cmpl $1, %ecx
; ENABLED-NEXT: jne .LBB3_4
; ENABLED-NEXT: # %bb.2: # %sw.bb.1
; ENABLED-NEXT: movl $111, (%rax)
; ENABLED-NEXT: movl $222, 4(%rax)
; ENABLED-NEXT: retq
; ENABLED-NEXT: .LBB3_3: # %sw.bb.2
; ENABLED-NEXT: movl $333, (%rax) # imm = 0x14D
; ENABLED-NEXT: movl $444, 4(%rax) # imm = 0x1BC
; ENABLED-NEXT: .LBB3_4: # %sw.epilog
; ENABLED-NEXT: retq
;
; DISABLED-LABEL: test4:
; DISABLED: # %bb.0: # %entry
; DISABLED-NEXT: imulq $12, %rdi, %rsi
; DISABLED-NEXT: movl arr1(%rsi), %edx
; DISABLED-NEXT: leaq arr1+4(%rsi), %rax
; DISABLED-NEXT: subl arr1+4(%rsi), %edx
; DISABLED-NEXT: leaq arr1+8(%rsi), %rcx
; DISABLED-NEXT: addl arr1+8(%rsi), %edx
; DISABLED-NEXT: cmpl $2, %edx
; DISABLED-NEXT: je .LBB3_3
; DISABLED-NEXT: # %bb.1: # %entry
; DISABLED-NEXT: cmpl $1, %edx
; DISABLED-NEXT: jne .LBB3_4
; DISABLED-NEXT: # %bb.2: # %sw.bb.1
; DISABLED-NEXT: movl $111, (%rax)
; DISABLED-NEXT: movl $222, (%rcx)
; DISABLED-NEXT: retq
; DISABLED-NEXT: .LBB3_3: # %sw.bb.2
; DISABLED-NEXT: movl $333, (%rax) # imm = 0x14D
; DISABLED-NEXT: movl $444, (%rcx) # imm = 0x1BC
; DISABLED-NEXT: .LBB3_4: # %sw.epilog
; DISABLED-NEXT: retq
LEA code size optimization pass (Part 2): Remove redundant LEA instructions. Make x86 OptimizeLEAs pass remove LEA instruction if there is another LEA (in the same basic block) which calculates address differing only be a displacement. Works only for -Oz. Differential Revision: http://reviews.llvm.org/D13295 llvm-svn: 257589
2016-01-13 19:30:44 +08:00
entry:
%a = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 0
%tmp = load i32, i32* %a, align 4
%b = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 1
%tmp1 = load i32, i32* %b, align 4
%sub = sub i32 %tmp, %tmp1
%c = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 2
%tmp2 = load i32, i32* %c, align 4
%add = add nsw i32 %sub, %tmp2
switch i32 %add, label %sw.epilog [
i32 1, label %sw.bb.1
i32 2, label %sw.bb.2
]
sw.bb.1: ; preds = %entry
store i32 111, i32* %b, align 4
store i32 222, i32* %c, align 4
br label %sw.epilog
sw.bb.2: ; preds = %entry
store i32 333, i32* %b, align 4
store i32 444, i32* %c, align 4
br label %sw.epilog
sw.epilog: ; preds = %sw.bb.2, %sw.bb.1, %entry
ret void
}