AMDGPU: Change boolean content type to 0 or 1

The usage of target boolean checks is overly inflexible, since sext
and zext of a compare are equally cheap. The choice is arbitrary, but
using 0/1 to some degree is the choice of lower resistance since
that's what most targets use. This enables a few combines that don't
bother to support ZeroOrNegativeOneBooleanContent.

9 files changed, 33 insertions, 21 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index cc8c36a93fe..f669c98969b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -451,9 +451,6 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(const TargetMachine &TM,
   for (int I = 0; I < RTLIB::UNKNOWN_LIBCALL; ++I)
     setLibcallName(static_cast<RTLIB::Libcall>(I), nullptr);
 
-  setBooleanContents(ZeroOrNegativeOneBooleanContent);
-  setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
-
   setSchedulingPreference(Sched::RegPressure);
   setJumpIsExpensive(true);
 
diff --git a/llvm/lib/Target/AMDGPU/R600ISelLowering.cpp b/llvm/lib/Target/AMDGPU/R600ISelLowering.cpp
index 697ea6607d5..6196be8ec83 100644
--- a/llvm/lib/Target/AMDGPU/R600ISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/R600ISelLowering.cpp
@@ -62,6 +62,9 @@ R600TargetLowering::R600TargetLowering(const TargetMachine &TM,
   addRegisterClass(MVT::v4f32, &R600::R600_Reg128RegClass);
   addRegisterClass(MVT::v4i32, &R600::R600_Reg128RegClass);
 
+  setBooleanContents(ZeroOrNegativeOneBooleanContent);
+  setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
+
   computeRegisterProperties(Subtarget->getRegisterInfo());
 
   // Legalize loads and stores to the private address space.
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index a11926ec2d7..e26f0e3b611 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -160,6 +160,13 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM,
 
   computeRegisterProperties(Subtarget->getRegisterInfo());
 
+  // The boolean content concept here is too inflexible. Compares only ever
+  // really produce a 1-bit result. Any copy/extend from these will turn into a
+  // select, and zext/1 or sext/-1 are equally cheap. Arbitrarily choose 0/1, as
+  // it's what most targets use.
+  setBooleanContents(ZeroOrOneBooleanContent);
+  setBooleanVectorContents(ZeroOrOneBooleanContent);
+
   // We need to custom lower vector stores from local memory
   setOperationAction(ISD::LOAD, MVT::v2i32, Custom);
   setOperationAction(ISD::LOAD, MVT::v3i32, Custom);
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index 3b0a41b9960..3737d0a7b41 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -542,7 +542,7 @@ void SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
       RC == &AMDGPU::SReg_32RegClass) {
     if (SrcReg == AMDGPU::SCC) {
       BuildMI(MBB, MI, DL, get(AMDGPU::S_CSELECT_B32), DestReg)
-          .addImm(-1)
+          .addImm(1)
           .addImm(0);
       return;
     }
@@ -840,7 +840,7 @@ void SIInstrInfo::insertVectorSelect(MachineBasicBlock &MBB,
       Register SReg = MRI.createVirtualRegister(BoolXExecRC);
       BuildMI(MBB, I, DL, get(ST.isWave32() ? AMDGPU::S_CSELECT_B32
                                             : AMDGPU::S_CSELECT_B64), SReg)
-        .addImm(-1)
+        .addImm(1)
         .addImm(0);
       BuildMI(MBB, I, DL, get(AMDGPU::V_CNDMASK_B32_e64), DstReg)
         .addImm(0)
@@ -855,7 +855,7 @@ void SIInstrInfo::insertVectorSelect(MachineBasicBlock &MBB,
       BuildMI(MBB, I, DL, get(ST.isWave32() ? AMDGPU::S_CSELECT_B32
                                             : AMDGPU::S_CSELECT_B64), SReg)
         .addImm(0)
-        .addImm(-1);
+        .addImm(1);
       BuildMI(MBB, I, DL, get(AMDGPU::V_CNDMASK_B32_e64), DstReg)
         .addImm(0)
         .addReg(FalseReg)
@@ -900,7 +900,7 @@ void SIInstrInfo::insertVectorSelect(MachineBasicBlock &MBB,
         .addImm(0);
       BuildMI(MBB, I, DL, get(ST.isWave32() ? AMDGPU::S_CSELECT_B32
                                             : AMDGPU::S_CSELECT_B64), SReg)
-        .addImm(-1)
+        .addImm(1)
         .addImm(0);
       BuildMI(MBB, I, DL, get(AMDGPU::V_CNDMASK_B32_e64), DstReg)
         .addImm(0)
@@ -919,7 +919,7 @@ void SIInstrInfo::insertVectorSelect(MachineBasicBlock &MBB,
       BuildMI(MBB, I, DL, get(ST.isWave32() ? AMDGPU::S_CSELECT_B32
                                             : AMDGPU::S_CSELECT_B64), SReg)
         .addImm(0)
-        .addImm(-1);
+        .addImm(1);
       BuildMI(MBB, I, DL, get(AMDGPU::V_CNDMASK_B32_e64), DstReg)
         .addImm(0)
         .addReg(FalseReg)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.if.break.i32.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.if.break.i32.ll
index 0f259fcb895..f135b67e43a 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.if.break.i32.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.if.break.i32.ll
@@ -9,7 +9,7 @@ define amdgpu_kernel void @test_wave32(i32 %arg0, [8 x i32], i32 %saved) {
 ; GCN-NEXT:    ; implicit-def: $vcc_hi
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_cmp_eq_u32 s0, 0
-; GCN-NEXT:    s_cselect_b32 s0, -1, 0
+; GCN-NEXT:    s_cselect_b32 s0, 1, 0
 ; GCN-NEXT:    v_cmp_ne_u32_e64 s0, 0, s0
 ; GCN-NEXT:    s_or_b32 s0, s0, s1
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.if.break.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.if.break.i64.ll
index e6d46f967cd..fc260b57ceb 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.if.break.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.if.break.i64.ll
@@ -8,7 +8,7 @@ define amdgpu_kernel void @test_wave64(i32 %arg0, [8 x i32], i64 %saved) {
 ; GCN-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0xa
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_cmp_eq_u32 s2, 0
-; GCN-NEXT:    s_cselect_b32 s2, -1, 0
+; GCN-NEXT:    s_cselect_b32 s2, 1, 0
 ; GCN-NEXT:    v_cmp_ne_u32_e64 s[2:3], 0, s2
 ; GCN-NEXT:    s_or_b64 s[0:1], s[2:3], s[0:1]
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
diff --git a/llvm/test/CodeGen/AMDGPU/amdgcn.private-memory.ll b/llvm/test/CodeGen/AMDGPU/amdgcn.private-memory.ll
index f93c96bc553..d38452a3a22 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgcn.private-memory.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgcn.private-memory.ll
@@ -13,7 +13,14 @@ declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
 
 ; GCN-LABEL: {{^}}work_item_info:
 ; GCN-NOT: v0
-; GCN: v_add_{{[iu]}}32_e32 [[RESULT:v[0-9]+]], vcc, v{{[0-9]+}}, v0
+; GCN: s_load_dword [[IN:s[0-9]+]]
+; GCN-NOT: v0
+
+; GCN-ALLOCA: v_add_{{[iu]}}32_e32 [[RESULT:v[0-9]+]], vcc, v{{[0-9]+}}, v0
+
+; GCN-PROMOTE: v_cmp_eq_u32_e64 vcc, [[IN]], 1
+; GCN-PROMOTE-NEXT: v_addc_u32_e32 [[RESULT:v[0-9]+]], vcc, 0, v0, vcc
+
 ; GCN: buffer_store_dword [[RESULT]]
 define amdgpu_kernel void @work_item_info(i32 addrspace(1)* %out, i32 %in) {
 entry:
diff --git a/llvm/test/CodeGen/AMDGPU/combine-cond-add-sub.ll b/llvm/test/CodeGen/AMDGPU/combine-cond-add-sub.ll
index 99d2bc9863f..37ae4d2d5c2 100644
--- a/llvm/test/CodeGen/AMDGPU/combine-cond-add-sub.ll
+++ b/llvm/test/CodeGen/AMDGPU/combine-cond-add-sub.ll
@@ -1,5 +1,5 @@
-; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
-; RUN: llc -march=amdgcn -mcpu=gfx902  -verify-machineinstrs < %s | FileCheck -check-prefix=GFX9 %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
+; RUN: llc -march=amdgcn -mcpu=gfx902  -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GFX9 %s
 
 ; GCN-LABEL: {{^}}add1:
 ; GCN: v_cmp_gt_u32_e{{32|64}} [[CC:[^,]+]], v{{[0-9]+}}, v{{[0-9]+}}
@@ -130,16 +130,14 @@ bb:
 ; GCN-LABEL: {{^}}sub_sube_commuted:
 ; GCN-DAG: v_cmp_gt_u32_e{{32|64}} [[CC:[^,]+]], v{{[0-9]+}}, v{{[0-9]+}}
 ; GCN-DAG: buffer_load_dword [[V:v[0-9]+]],
-; GCN:     v_subbrev_u32_e{{32|64}} [[SUBB:v[0-9]+]], {{[^,]+}}, 0, [[V]], [[CC]]
-; GCN:     v_sub_i32_e32 [[SUB:v[0-9]+]], vcc, s{{[0-9]+}}, [[SUBB]]
-; GCN:     v_add_i32_e32 {{.*}}, 0x64, [[SUB]]
+; GCN: v_addc_u32_e32 [[ADDC:v[0-9]+]], vcc, v{{[0-9]+}}, v{{[0-9]+}}, [[CC]]
+; GCN: v_add_i32_e32 {{.*}}, 0x64, [[ADDC]]
 
 ; GFX9-LABEL: {{^}}sub_sube_commuted:
 ; GFX9-DAG: v_cmp_gt_u32_e{{32|64}} [[CC:[^,]+]], v{{[0-9]+}}, v{{[0-9]+}}
 ; GFX9-DAG: global_load_dword [[V:v[0-9]+]],
-; GFX9:     v_subbrev_co_u32_e{{32|64}} [[SUBB:v[0-9]+]], {{[^,]+}}, 0, [[V]], [[CC]]
-; GFX9:     v_sub_u32_e32 [[SUB:v[0-9]+]], s{{[0-9]+}}, [[SUBB]]
-; GFX9:     v_add_u32_e32 {{.*}}, 0x64, [[SUB]]
+; GFX9: v_addc_co_u32_e32 [[ADDC:v[0-9]+]], vcc, v{{[0-9]+}}, v{{[0-9]+}}, [[CC]]
+; GFX9: v_add_u32_e32 {{.*}}, 0x64, [[ADDC]]
 define amdgpu_kernel void @sub_sube_commuted(i32 addrspace(1)* nocapture %arg, i32 %a) {
 bb:
   %x = tail call i32 @llvm.amdgcn.workitem.id.x()
diff --git a/llvm/test/CodeGen/AMDGPU/extract_vector_dynelt.ll b/llvm/test/CodeGen/AMDGPU/extract_vector_dynelt.ll
index a2b68cc4214..eb95efd1860 100644
--- a/llvm/test/CodeGen/AMDGPU/extract_vector_dynelt.ll
+++ b/llvm/test/CodeGen/AMDGPU/extract_vector_dynelt.ll
@@ -306,10 +306,10 @@ entry:
 
 ; GCN-LABEL: {{^}}bit128_extelt:
 ; GCN-NOT: buffer_
-; GCN-DAG: v_cndmask_b32_e{{32|64}} [[V1:v[0-9]+]], 1, 0,
+; GCN-DAG: v_cndmask_b32_e{{32|64}} [[V1:v[0-9]+]], 0, 1
 ; GCN-DAG: v_mov_b32_e32 [[LASTIDX:v[0-9]+]], 0x7f
 ; GCN-DAG: v_cmp_ne_u32_e32 [[CL:[^,]+]], s{{[0-9]+}}, [[LASTIDX]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} [[VL:v[0-9]+]], 0, v{{[0-9]+}}, [[CL]]
+; GCN-DAG: v_cndmask_b32_e{{32|64}} [[VL:v[0-9]+]], 0, [[V1]], [[CL]]
 ; GCN:     v_and_b32_e32 [[RES:v[0-9]+]], 1, [[VL]]
 ; GCN:     store_dword v[{{[0-9:]+}}], [[RES]]
 define amdgpu_kernel void @bit128_extelt(i32 addrspace(1)* %out, i32 %sel) {