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/AMDGPU/commute-compares.ll

718 lines
28 KiB
LLVM
Raw Normal View History

R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
declare i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
; --------------------------------------------------------------------------------
; i32 compares
; --------------------------------------------------------------------------------
; GCN-LABEL: {{^}}commute_eq_64_i32:
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 09:50:20 +08:00
; GCN: v_cmp_eq_u32_e32 vcc, 64, v{{[0-9]+}}
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
define void @commute_eq_64_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp eq i32 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ne_64_i32:
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 09:50:20 +08:00
; GCN: v_cmp_ne_u32_e32 vcc, 64, v{{[0-9]+}}
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
define void @commute_ne_64_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp ne i32 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; FIXME: Why isn't this being folded as a constant?
; GCN-LABEL: {{^}}commute_ne_litk_i32:
; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 0x3039
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 09:50:20 +08:00
; GCN: v_cmp_ne_u32_e32 vcc, [[K]], v{{[0-9]+}}
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
define void @commute_ne_litk_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp ne i32 %val, 12345
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ugt_64_i32:
; GCN: v_cmp_lt_u32_e32 vcc, 64, v{{[0-9]+}}
define void @commute_ugt_64_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp ugt i32 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_uge_64_i32:
; GCN: v_cmp_lt_u32_e32 vcc, 63, v{{[0-9]+}}
define void @commute_uge_64_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp uge i32 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ult_64_i32:
; GCN: v_cmp_gt_u32_e32 vcc, 64, v{{[0-9]+}}
define void @commute_ult_64_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp ult i32 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ule_63_i32:
; GCN: v_cmp_gt_u32_e32 vcc, 64, v{{[0-9]+}}
define void @commute_ule_63_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp ule i32 %val, 63
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; FIXME: Undo canonicalization to gt (x + 1) since it doesn't use the inline imm
; GCN-LABEL: {{^}}commute_ule_64_i32:
; GCN: v_mov_b32_e32 [[K:v[0-9]+]], 0x41{{$}}
; GCN: v_cmp_gt_u32_e32 vcc, [[K]], v{{[0-9]+}}
define void @commute_ule_64_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp ule i32 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_sgt_neg1_i32:
; GCN: v_cmp_lt_i32_e32 vcc, -1, v{{[0-9]+}}
define void @commute_sgt_neg1_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp sgt i32 %val, -1
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_sge_neg2_i32:
; GCN: v_cmp_lt_i32_e32 vcc, -3, v{{[0-9]+}}
define void @commute_sge_neg2_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp sge i32 %val, -2
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_slt_neg16_i32:
; GCN: v_cmp_gt_i32_e32 vcc, -16, v{{[0-9]+}}
define void @commute_slt_neg16_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp slt i32 %val, -16
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_sle_5_i32:
; GCN: v_cmp_gt_i32_e32 vcc, 6, v{{[0-9]+}}
define void @commute_sle_5_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i32, i32 addrspace(1)* %gep.in
%cmp = icmp sle i32 %val, 5
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; --------------------------------------------------------------------------------
; i64 compares
; --------------------------------------------------------------------------------
; GCN-LABEL: {{^}}commute_eq_64_i64:
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 09:50:20 +08:00
; GCN: v_cmp_eq_u64_e32 vcc, 64, v{{\[[0-9]+:[0-9]+\]}}
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
define void @commute_eq_64_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp eq i64 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ne_64_i64:
AMDGPU: Use unsigned compare for eq/ne For some reason there are both of these available, except for scalar 64-bit compares which only has u64. I'm not sure why there are both (I'm guessing it's for the one bit inputs we don't use), but for consistency always using the unsigned one. llvm-svn: 282832
2016-09-30 09:50:20 +08:00
; GCN: v_cmp_ne_u64_e32 vcc, 64, v{{\[[0-9]+:[0-9]+\]}}
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
define void @commute_ne_64_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp ne i64 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ugt_64_i64:
; GCN: v_cmp_lt_u64_e32 vcc, 64, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_ugt_64_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp ugt i64 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_uge_64_i64:
; GCN: v_cmp_lt_u64_e32 vcc, 63, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_uge_64_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp uge i64 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ult_64_i64:
; GCN: v_cmp_gt_u64_e32 vcc, 64, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_ult_64_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp ult i64 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ule_63_i64:
; GCN: v_cmp_gt_u64_e32 vcc, 64, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_ule_63_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp ule i64 %val, 63
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; FIXME: Undo canonicalization to gt (x + 1) since it doesn't use the inline imm
; GCN-LABEL: {{^}}commute_ule_64_i64:
; GCN-DAG: s_movk_i32 s[[KLO:[0-9]+]], 0x41{{$}}
; GCN: v_cmp_gt_u64_e32 vcc, s{{\[}}[[KLO]]:{{[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_ule_64_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp ule i64 %val, 64
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_sgt_neg1_i64:
; GCN: v_cmp_lt_i64_e32 vcc, -1, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_sgt_neg1_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp sgt i64 %val, -1
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_sge_neg2_i64:
; GCN: v_cmp_lt_i64_e32 vcc, -3, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_sge_neg2_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp sge i64 %val, -2
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_slt_neg16_i64:
; GCN: v_cmp_gt_i64_e32 vcc, -16, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_slt_neg16_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp slt i64 %val, -16
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_sle_5_i64:
; GCN: v_cmp_gt_i64_e32 vcc, 6, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_sle_5_i64(i32 addrspace(1)* %out, i64 addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load i64, i64 addrspace(1)* %gep.in
%cmp = icmp sle i64 %val, 5
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; --------------------------------------------------------------------------------
; f32 compares
; --------------------------------------------------------------------------------
; GCN-LABEL: {{^}}commute_oeq_2.0_f32:
; GCN: v_cmp_eq_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_oeq_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp oeq float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ogt_2.0_f32:
; GCN: v_cmp_lt_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_ogt_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp ogt float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_oge_2.0_f32:
; GCN: v_cmp_le_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_oge_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp oge float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_olt_2.0_f32:
; GCN: v_cmp_gt_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_olt_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp olt float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ole_2.0_f32:
; GCN: v_cmp_ge_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_ole_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp ole float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_one_2.0_f32:
; GCN: v_cmp_lg_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_one_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp one float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ord_2.0_f32:
; GCN: v_cmp_o_f32_e32 vcc, [[REG:v[0-9]+]], [[REG]]
define void @commute_ord_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp ord float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ueq_2.0_f32:
; GCN: v_cmp_nlg_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_ueq_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp ueq float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ugt_2.0_f32:
; GCN: v_cmp_nge_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_ugt_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp ugt float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_uge_2.0_f32:
; GCN: v_cmp_ngt_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_uge_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp uge float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ult_2.0_f32:
; GCN: v_cmp_nle_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_ult_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp ult float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ule_2.0_f32:
; GCN: v_cmp_nlt_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_ule_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp ule float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_une_2.0_f32:
; GCN: v_cmp_neq_f32_e32 vcc, 2.0, v{{[0-9]+}}
define void @commute_une_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp une float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_uno_2.0_f32:
; GCN: v_cmp_u_f32_e32 vcc, [[REG:v[0-9]+]], [[REG]]
define void @commute_uno_2.0_f32(i32 addrspace(1)* %out, float addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr float, float addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load float, float addrspace(1)* %gep.in
%cmp = fcmp uno float %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; --------------------------------------------------------------------------------
; f64 compares
; --------------------------------------------------------------------------------
; GCN-LABEL: {{^}}commute_oeq_2.0_f64:
; GCN: v_cmp_eq_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_oeq_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp oeq double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ogt_2.0_f64:
; GCN: v_cmp_lt_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_ogt_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp ogt double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_oge_2.0_f64:
; GCN: v_cmp_le_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_oge_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp oge double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_olt_2.0_f64:
; GCN: v_cmp_gt_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_olt_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp olt double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ole_2.0_f64:
; GCN: v_cmp_ge_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_ole_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp ole double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_one_2.0_f64:
; GCN: v_cmp_lg_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_one_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp one double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ord_2.0_f64:
; GCN: v_cmp_o_f64_e32 vcc, [[REG:v\[[0-9]+:[0-9]+\]]], [[REG]]
define void @commute_ord_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp ord double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ueq_2.0_f64:
; GCN: v_cmp_nlg_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_ueq_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp ueq double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ugt_2.0_f64:
; GCN: v_cmp_nge_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_ugt_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp ugt double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_uge_2.0_f64:
; GCN: v_cmp_ngt_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_uge_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp uge double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ult_2.0_f64:
; GCN: v_cmp_nle_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_ult_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp ult double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_ule_2.0_f64:
; GCN: v_cmp_nlt_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_ule_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp ule double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_une_2.0_f64:
; GCN: v_cmp_neq_f64_e32 vcc, 2.0, v{{\[[0-9]+:[0-9]+\]}}
define void @commute_une_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp une double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}commute_uno_2.0_f64:
; GCN: v_cmp_u_f64_e32 vcc, [[REG:v\[[0-9]+:[0-9]+\]]], [[REG]]
define void @commute_uno_2.0_f64(i32 addrspace(1)* %out, double addrspace(1)* %in) #1 {
AMDGPU: Remove some old intrinsic uses from tests llvm-svn: 260493
2016-02-11 14:02:01 +08:00
%tid = call i32 @llvm.amdgcn.workitem.id.x() #0
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
%gep.in = getelementptr double, double addrspace(1)* %in, i32 %tid
%gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
%val = load double, double addrspace(1)* %gep.in
%cmp = fcmp uno double %val, 2.0
%ext = sext i1 %cmp to i32
store i32 %ext, i32 addrspace(1)* %gep.out
ret void
}
AMDGPU: Fix handling of 16-bit immediates Since 32-bit instructions with 32-bit input immediate behavior are used to materialize 16-bit constants in 32-bit registers for 16-bit instructions, determining the legality based on the size is incorrect. Change operands to have the size specified in the type. Also adds a workaround for a disassembler bug that produces an immediate MCOperand for an operand that is supposed to be OPERAND_REGISTER. The assembler appears to accept out of bounds immediates and truncates them, but this seems to be an issue for 32-bit already. llvm-svn: 289306
2016-12-10 08:39:12 +08:00
; FIXME: Should be able to fold this frameindex
AMDGPU: Support commuting a FrameIndex operand llvm-svn: 281369
2016-09-14 03:03:12 +08:00
; Without commuting the frame index in the pre-regalloc run of
; SIShrinkInstructions, this was using the VOP3 compare.
; GCN-LABEL: {{^}}commute_frameindex:
AMDGPU: Fix handling of 16-bit immediates Since 32-bit instructions with 32-bit input immediate behavior are used to materialize 16-bit constants in 32-bit registers for 16-bit instructions, determining the legality based on the size is incorrect. Change operands to have the size specified in the type. Also adds a workaround for a disassembler bug that produces an immediate MCOperand for an operand that is supposed to be OPERAND_REGISTER. The assembler appears to accept out of bounds immediates and truncates them, but this seems to be an issue for 32-bit already. llvm-svn: 289306
2016-12-10 08:39:12 +08:00
; XGCN: v_cmp_eq_u32_e32 vcc, 0, v{{[0-9]+}}
; GCN: v_mov_b32_e32 [[FI:v[0-9]+]], 0{{$}}
; GCN: v_cmp_eq_u32_e32 vcc, [[FI]], v{{[0-9]+}}
AMDGPU: Support commuting a FrameIndex operand llvm-svn: 281369
2016-09-14 03:03:12 +08:00
define void @commute_frameindex(i32 addrspace(1)* nocapture %out) #0 {
entry:
%stack0 = alloca i32
%ptr0 = load volatile i32*, i32* addrspace(1)* undef
%eq = icmp eq i32* %ptr0, %stack0
%ext = zext i1 %eq to i32
store volatile i32 %ext, i32 addrspace(1)* %out
ret void
}
R600/SI: Allow commuting compares This enables very common cases to switch to the smaller encoding. All of the standard LLVM canonicalizations of comparisons are the opposite of what we want. Compares with constants are moved to the RHS, but the first operand can be an inline immediate, literal constant, or SGPR using the 32-bit VOPC encoding. There are additional bad canonicalizations that should also be fixed, such as canonicalizing ge x, k to gt x, (k + 1) if this makes k no longer an inline immediate value. llvm-svn: 232988
2015-03-24 02:45:30 +08:00
attributes #0 = { nounwind readnone }
attributes #1 = { nounwind }