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/gpr-to-mask.ll

359 lines
10 KiB
LLVM
Raw Normal View History

[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512dq < %s | FileCheck %s
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
define void @test_fcmp_storefloat(i1 %cond, float* %fptr, float %f1, float %f2, float %f3, float %f4, float %f5, float %f6) {
; CHECK-LABEL: test_fcmp_storefloat:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: testb $1, %dil
; CHECK-NEXT: je .LBB0_2
; CHECK-NEXT: # BB#1: # %if
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: vcmpeqss %xmm3, %xmm2, %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: jmp .LBB0_3
; CHECK-NEXT: .LBB0_2: # %else
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: vcmpeqss %xmm5, %xmm4, %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: .LBB0_3: # %exit
; CHECK-NEXT: vmovss %xmm0, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vmovss %xmm1, (%rsi)
; CHECK-NEXT: retq
entry:
br i1 %cond, label %if, label %else
if:
%cmp1 = fcmp oeq float %f3, %f4
br label %exit
else:
%cmp2 = fcmp oeq float %f5, %f6
br label %exit
exit:
%val = phi i1 [%cmp1, %if], [%cmp2, %else]
%selected = select i1 %val, float %f1, float %f2
store float %selected, float* %fptr
ret void
}
define void @test_fcmp_storei1(i1 %cond, float* %fptr, i1* %iptr, float %f1, float %f2, float %f3, float %f4) {
; CHECK-LABEL: test_fcmp_storei1:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: testb $1, %dil
; CHECK-NEXT: je .LBB1_2
; CHECK-NEXT: # BB#1: # %if
; CHECK-NEXT: vcmpeqss %xmm1, %xmm0, %k0
; CHECK-NEXT: jmp .LBB1_3
; CHECK-NEXT: .LBB1_2: # %else
; CHECK-NEXT: vcmpeqss %xmm3, %xmm2, %k0
; CHECK-NEXT: .LBB1_3: # %exit
; CHECK-NEXT: kmovd %k0, %eax
; CHECK-NEXT: andb $1, %al
; CHECK-NEXT: movb %al, (%rdx)
; CHECK-NEXT: retq
entry:
br i1 %cond, label %if, label %else
if:
%cmp1 = fcmp oeq float %f1, %f2
br label %exit
else:
%cmp2 = fcmp oeq float %f3, %f4
br label %exit
exit:
%val = phi i1 [%cmp1, %if], [%cmp2, %else]
store i1 %val, i1* %iptr
ret void
}
define void @test_load_add(i1 %cond, float* %fptr, i1* %iptr1, i1* %iptr2, float %f1, float %f2) {
; CHECK-LABEL: test_load_add:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: testb $1, %dil
; CHECK-NEXT: je .LBB2_2
; CHECK-NEXT: # BB#1: # %if
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rdx), %k0
; CHECK-NEXT: kmovb (%rcx), %k1
; CHECK-NEXT: kaddb %k1, %k0, %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: jmp .LBB2_3
; CHECK-NEXT: .LBB2_2: # %else
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rcx), %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: .LBB2_3: # %exit
; CHECK-NEXT: vmovss %xmm0, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vmovss %xmm1, (%rsi)
; CHECK-NEXT: retq
entry:
br i1 %cond, label %if, label %else
if:
%loaded1 = load i1, i1* %iptr1
%loaded2if = load i1, i1* %iptr2
%added = add i1 %loaded1, %loaded2if
br label %exit
else:
%loaded2else = load i1, i1* %iptr2
br label %exit
exit:
%val = phi i1 [%added, %if], [%loaded2else, %else]
%selected = select i1 %val, float %f1, float %f2
store float %selected, float* %fptr
ret void
}
define void @test_load_i1(i1 %cond, float* %fptr, i1* %iptr1, i1* %iptr2, float %f1, float %f2) {
; CHECK-LABEL: test_load_i1:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: testb $1, %dil
; CHECK-NEXT: je .LBB3_2
; CHECK-NEXT: # BB#1: # %if
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rdx), %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: jmp .LBB3_3
; CHECK-NEXT: .LBB3_2: # %else
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rcx), %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: .LBB3_3: # %exit
; CHECK-NEXT: vmovss %xmm0, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vmovss %xmm1, (%rsi)
; CHECK-NEXT: retq
entry:
br i1 %cond, label %if, label %else
if:
%loaded1 = load i1, i1* %iptr1
br label %exit
else:
%loaded2 = load i1, i1* %iptr2
br label %exit
exit:
%val = phi i1 [%loaded1, %if], [%loaded2, %else]
%selected = select i1 %val, float %f1, float %f2
store float %selected, float* %fptr
ret void
}
define void @test_loadi1_storei1(i1 %cond, i1* %iptr1, i1* %iptr2, i1* %iptr3) {
; CHECK-LABEL: test_loadi1_storei1:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: testb $1, %dil
; CHECK-NEXT: je .LBB4_2
; CHECK-NEXT: # BB#1: # %if
; CHECK-NEXT: movb (%rsi), %al
; CHECK-NEXT: jmp .LBB4_3
; CHECK-NEXT: .LBB4_2: # %else
; CHECK-NEXT: movb (%rdx), %al
; CHECK-NEXT: .LBB4_3: # %exit
; CHECK-NEXT: andb $1, %al
; CHECK-NEXT: movb %al, (%rcx)
; CHECK-NEXT: retq
entry:
br i1 %cond, label %if, label %else
if:
%loaded1 = load i1, i1* %iptr1
br label %exit
else:
%loaded2 = load i1, i1* %iptr2
br label %exit
exit:
%val = phi i1 [%loaded1, %if], [%loaded2, %else]
store i1 %val, i1* %iptr3
ret void
}
define void @test_shl1(i1 %cond, i8* %ptr1, i8* %ptr2, <8 x float> %fvec1, <8 x float> %fvec2, <8 x float>* %fptrvec) {
; CHECK-LABEL: test_shl1:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: # kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
; CHECK-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
; CHECK-NEXT: testb $1, %dil
; CHECK-NEXT: je .LBB5_2
; CHECK-NEXT: # BB#1: # %if
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rsi), %k0
; CHECK-NEXT: kaddb %k0, %k0, %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: jmp .LBB5_3
; CHECK-NEXT: .LBB5_2: # %else
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rdx), %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: .LBB5_3: # %exit
; CHECK-NEXT: vmovaps %zmm0, %zmm1 {%k1}
; CHECK-NEXT: vmovaps %ymm1, (%rcx)
; CHECK-NEXT: vzeroupper
; CHECK-NEXT: retq
entry:
br i1 %cond, label %if, label %else
if:
%loaded1 = load i8, i8* %ptr1
%shifted = shl i8 %loaded1, 1
br label %exit
else:
%loaded2 = load i8, i8* %ptr2
br label %exit
exit:
%val = phi i8 [%shifted, %if], [%loaded2, %else]
%mask = bitcast i8 %val to <8 x i1>
%selected = select <8 x i1> %mask, <8 x float> %fvec1, <8 x float> %fvec2
store <8 x float> %selected, <8 x float>* %fptrvec
ret void
}
define void @test_shr1(i1 %cond, i8* %ptr1, i8* %ptr2, <8 x float> %fvec1, <8 x float> %fvec2, <8 x float>* %fptrvec) {
; CHECK-LABEL: test_shr1:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: # kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
; CHECK-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
; CHECK-NEXT: testb $1, %dil
; CHECK-NEXT: je .LBB6_2
; CHECK-NEXT: # BB#1: # %if
; CHECK-NEXT: movb (%rsi), %al
; CHECK-NEXT: shrb %al
; CHECK-NEXT: jmp .LBB6_3
; CHECK-NEXT: .LBB6_2: # %else
; CHECK-NEXT: movb (%rdx), %al
; CHECK-NEXT: .LBB6_3: # %exit
; CHECK-NEXT: kmovd %eax, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm1 {%k1}
; CHECK-NEXT: vmovaps %ymm1, (%rcx)
; CHECK-NEXT: vzeroupper
; CHECK-NEXT: retq
entry:
br i1 %cond, label %if, label %else
if:
%loaded1 = load i8, i8* %ptr1
%shifted = lshr i8 %loaded1, 1
br label %exit
else:
%loaded2 = load i8, i8* %ptr2
br label %exit
exit:
%val = phi i8 [%shifted, %if], [%loaded2, %else]
%mask = bitcast i8 %val to <8 x i1>
%selected = select <8 x i1> %mask, <8 x float> %fvec1, <8 x float> %fvec2
store <8 x float> %selected, <8 x float>* %fptrvec
ret void
}
define void @test_shr2(i1 %cond, i8* %ptr1, i8* %ptr2, <8 x float> %fvec1, <8 x float> %fvec2, <8 x float>* %fptrvec) {
; CHECK-LABEL: test_shr2:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: # kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
; CHECK-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
; CHECK-NEXT: testb $1, %dil
; CHECK-NEXT: je .LBB7_2
; CHECK-NEXT: # BB#1: # %if
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rsi), %k0
; CHECK-NEXT: kshiftrb $2, %k0, %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: jmp .LBB7_3
; CHECK-NEXT: .LBB7_2: # %else
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rdx), %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: .LBB7_3: # %exit
; CHECK-NEXT: vmovaps %zmm0, %zmm1 {%k1}
; CHECK-NEXT: vmovaps %ymm1, (%rcx)
; CHECK-NEXT: vzeroupper
; CHECK-NEXT: retq
entry:
br i1 %cond, label %if, label %else
if:
%loaded1 = load i8, i8* %ptr1
%shifted = lshr i8 %loaded1, 2
br label %exit
else:
%loaded2 = load i8, i8* %ptr2
br label %exit
exit:
%val = phi i8 [%shifted, %if], [%loaded2, %else]
%mask = bitcast i8 %val to <8 x i1>
%selected = select <8 x i1> %mask, <8 x float> %fvec1, <8 x float> %fvec2
store <8 x float> %selected, <8 x float>* %fptrvec
ret void
}
define void @test_shl(i1 %cond, i8* %ptr1, i8* %ptr2, <8 x float> %fvec1, <8 x float> %fvec2, <8 x float>* %fptrvec) {
; CHECK-LABEL: test_shl:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: # kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
; CHECK-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
; CHECK-NEXT: testb $1, %dil
; CHECK-NEXT: je .LBB8_2
; CHECK-NEXT: # BB#1: # %if
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rsi), %k0
; CHECK-NEXT: kshiftlb $6, %k0, %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: jmp .LBB8_3
; CHECK-NEXT: .LBB8_2: # %else
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rdx), %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: .LBB8_3: # %exit
; CHECK-NEXT: vmovaps %zmm0, %zmm1 {%k1}
; CHECK-NEXT: vmovaps %ymm1, (%rcx)
; CHECK-NEXT: vzeroupper
; CHECK-NEXT: retq
entry:
br i1 %cond, label %if, label %else
if:
%loaded1 = load i8, i8* %ptr1
%shifted = shl i8 %loaded1, 6
br label %exit
else:
%loaded2 = load i8, i8* %ptr2
br label %exit
exit:
%val = phi i8 [%shifted, %if], [%loaded2, %else]
%mask = bitcast i8 %val to <8 x i1>
%selected = select <8 x i1> %mask, <8 x float> %fvec1, <8 x float> %fvec2
store <8 x float> %selected, <8 x float>* %fptrvec
ret void
}
define void @test_add(i1 %cond, i8* %ptr1, i8* %ptr2, <8 x float> %fvec1, <8 x float> %fvec2, <8 x float>* %fptrvec) {
; CHECK-LABEL: test_add:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: # kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
; CHECK-NEXT: # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kmovb (%rsi), %k0
; CHECK-NEXT: kmovb (%rdx), %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: testb $1, %dil
; CHECK-NEXT: je .LBB9_2
; CHECK-NEXT: # BB#1: # %if
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kandb %k1, %k0, %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: jmp .LBB9_3
; CHECK-NEXT: .LBB9_2: # %else
[X86] Add a pass to convert instruction chains between domains. The pass scans the function to find instruction chains that define registers in the same domain (closures). It then calculates the cost of converting the closure to another domain. If found profitable, the instructions are converted to instructions in the other domain and the register classes are changed accordingly. This commit adds the pass infrastructure and a simple conversion from the GPR domain to the Mask domain. Differential Revision: https://reviews.llvm.org/D37251 Change-Id: Ic2cf1d76598110401168326d411128ae2580a604 llvm-svn: 316288
2017-10-22 19:43:08 +08:00
; CHECK-NEXT: kaddb %k1, %k0, %k1
[X86] Add a test cases to demonstrate selecting GPR instructions when using mask based ones are more appropriate. llvm-svn: 311996
2017-08-29 19:58:03 +08:00
; CHECK-NEXT: .LBB9_3: # %exit
; CHECK-NEXT: vmovaps %zmm0, %zmm1 {%k1}
; CHECK-NEXT: vmovaps %ymm1, (%rcx)
; CHECK-NEXT: vzeroupper
; CHECK-NEXT: retq
entry:
%loaded1 = load i8, i8* %ptr1
%loaded2 = load i8, i8* %ptr2
br i1 %cond, label %if, label %else
if:
%and = and i8 %loaded1, %loaded2
br label %exit
else:
%add = add i8 %loaded1, %loaded2
br label %exit
exit:
%val = phi i8 [%and, %if], [%add, %else]
%mask = bitcast i8 %val to <8 x i1>
%selected = select <8 x i1> %mask, <8 x float> %fvec1, <8 x float> %fvec2
store <8 x float> %selected, <8 x float>* %fptrvec
ret void
}