[llvm] [RISCV] Add register bank and instruction selection support for FP G_SELECT. (PR #72726)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Nov 17 16:41:20 PST 2023
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-globalisel
Author: Craig Topper (topperc)
<details>
<summary>Changes</summary>
Try to pick the FP register bank based on surrounding use/defs. Code is basically copied from AArch64.
Need legalizer changes to make this more useful. Right now we're stuck with only being able to FP select types the same size as XLen.
---
Full diff: https://github.com/llvm/llvm-project/pull/72726.diff
6 Files Affected:
- (modified) llvm/lib/Target/RISCV/GISel/RISCVInstructionSelector.cpp (+10-2)
- (modified) llvm/lib/Target/RISCV/GISel/RISCVRegisterBankInfo.cpp (+47-4)
- (added) llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-select-rv32.mir (+33)
- (added) llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-select-rv64.mir (+34)
- (added) llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-select-rv32.mir (+34)
- (added) llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-select-rv64.mir (+34)
``````````diff
diff --git a/llvm/lib/Target/RISCV/GISel/RISCVInstructionSelector.cpp b/llvm/lib/Target/RISCV/GISel/RISCVInstructionSelector.cpp
index 25f4a217c070349..2586390e8562d1c 100644
--- a/llvm/lib/Target/RISCV/GISel/RISCVInstructionSelector.cpp
+++ b/llvm/lib/Target/RISCV/GISel/RISCVInstructionSelector.cpp
@@ -816,8 +816,16 @@ bool RISCVInstructionSelector::selectSelect(MachineInstr &MI,
RISCVCC::CondCode CC;
getOperandsForBranch(SelectMI.getCondReg(), MRI, CC, LHS, RHS);
- MachineInstr *Result = MIB.buildInstr(RISCV::Select_GPR_Using_CC_GPR)
- .addDef(SelectMI.getReg(0))
+ Register DstReg = SelectMI.getReg(0);
+
+ unsigned Opc = RISCV::Select_GPR_Using_CC_GPR;
+ if (RBI.getRegBank(DstReg, MRI, TRI)->getID() == RISCV::FPRBRegBankID) {
+ unsigned Size = MRI.getType(DstReg).getSizeInBits();
+ Opc = Size == 32 ? RISCV::Select_FPR32_Using_CC_GPR : RISCV::Select_FPR64_Using_CC_GPR;
+ }
+
+ MachineInstr *Result = MIB.buildInstr(Opc)
+ .addDef(DstReg)
.addReg(LHS)
.addReg(RHS)
.addImm(CC)
diff --git a/llvm/lib/Target/RISCV/GISel/RISCVRegisterBankInfo.cpp b/llvm/lib/Target/RISCV/GISel/RISCVRegisterBankInfo.cpp
index 6d77e2b7edd9010..5721bd9609040e6 100644
--- a/llvm/lib/Target/RISCV/GISel/RISCVRegisterBankInfo.cpp
+++ b/llvm/lib/Target/RISCV/GISel/RISCVRegisterBankInfo.cpp
@@ -323,12 +323,55 @@ RISCVRegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
OpdsMapping[0] = getFPValueMapping(Ty.getSizeInBits());
break;
}
- case TargetOpcode::G_SELECT:
- OpdsMapping[0] = GPRValueMapping;
+ case TargetOpcode::G_SELECT: {
+ LLT Ty = MRI.getType(MI.getOperand(0).getReg());
+
+ // Try to minimize the number of copies. If we have more floating point
+ // constrained values than not, then we'll put everything on FPR. Otherwise,
+ // everything has to be on GPR.
+ unsigned NumFP = 0;
+
+ // Check if the uses of the result always produce floating point values.
+ //
+ // For example:
+ //
+ // %z = G_SELECT %cond %x %y
+ // fpr = G_FOO %z ...
+ if (any_of(MRI.use_nodbg_instructions(MI.getOperand(0).getReg()),
+ [&](const MachineInstr &UseMI) { return onlyUsesFP(UseMI, MRI, TRI); }))
+ ++NumFP;
+
+ // Check if the defs of the source values always produce floating point
+ // values.
+ //
+ // For example:
+ //
+ // %x = G_SOMETHING_ALWAYS_FLOAT %a ...
+ // %z = G_SELECT %cond %x %y
+ //
+ // Also check whether or not the sources have already been decided to be
+ // FPR. Keep track of this.
+ //
+ // This doesn't check the condition, since it's just whatever is in NZCV.
+ // This isn't passed explicitly in a register to fcsel/csel.
+ for (unsigned Idx = 2; Idx < 4; ++Idx) {
+ Register VReg = MI.getOperand(Idx).getReg();
+ MachineInstr *DefMI = MRI.getVRegDef(VReg);
+ if (getRegBank(VReg, MRI, TRI) == &RISCV::FPRBRegBank ||
+ onlyDefinesFP(*DefMI, MRI, TRI))
+ ++NumFP;
+ }
+
+ // Condition operand is always GPR.
OpdsMapping[1] = GPRValueMapping;
- OpdsMapping[2] = GPRValueMapping;
- OpdsMapping[3] = GPRValueMapping;
+
+ const ValueMapping *Mapping = GPRValueMapping;
+ if (NumFP >= 2)
+ Mapping = getFPValueMapping(Ty.getSizeInBits());
+
+ OpdsMapping[0] = OpdsMapping[2] = OpdsMapping[3] = Mapping;
break;
+ }
case TargetOpcode::G_FPTOSI:
case TargetOpcode::G_FPTOUI: {
LLT Ty = MRI.getType(MI.getOperand(1).getReg());
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-select-rv32.mir b/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-select-rv32.mir
new file mode 100644
index 000000000000000..4fffbc42b2cff34
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-select-rv32.mir
@@ -0,0 +1,33 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=riscv32 -mattr=+d -run-pass=instruction-select \
+# RUN: -simplify-mir -verify-machineinstrs %s -o - | FileCheck %s
+
+---
+name: fp_select_s32
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $f10_f, $f11_f
+
+ ; CHECK-LABEL: name: fp_select_s32
+ ; CHECK: liveins: $x10, $f10_f, $f11_f
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr = COPY $x10
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fpr32 = COPY $f10_f
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:fpr32 = COPY $f11_f
+ ; CHECK-NEXT: [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY]], 1
+ ; CHECK-NEXT: [[Select_FPR32_Using_CC_GPR:%[0-9]+]]:fpr32 = Select_FPR32_Using_CC_GPR [[ANDI]], $x0, 1, [[COPY1]], [[COPY2]]
+ ; CHECK-NEXT: $f10_f = COPY [[Select_FPR32_Using_CC_GPR]]
+ ; CHECK-NEXT: PseudoRET implicit $f10_f
+ %0:gprb(s32) = COPY $x10
+ %1:fprb(s32) = COPY $f10_f
+ %2:fprb(s32) = COPY $f11_f
+ %3:gprb(s32) = G_CONSTANT i32 1
+ %4:gprb(s32) = G_AND %0, %3
+ %5:fprb(s32) = G_SELECT %4(s32), %1, %2
+ $f10_f = COPY %5(s32)
+ PseudoRET implicit $f10_f
+
+...
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-select-rv64.mir b/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-select-rv64.mir
new file mode 100644
index 000000000000000..8efb8b10d8a931d
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-select-rv64.mir
@@ -0,0 +1,34 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=riscv64 -mattr=+d -run-pass=instruction-select \
+# RUN: -simplify-mir -verify-machineinstrs %s -o - | FileCheck %s
+
+---
+name: fp_select_s64
+alignment: 1
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $f10_d, $f11_d
+
+ ; CHECK-LABEL: name: fp_select_s64
+ ; CHECK: liveins: $x10, $f10_d, $f11_d
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr = COPY $x10
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fpr64 = COPY $f10_d
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:fpr64 = COPY $f11_d
+ ; CHECK-NEXT: [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY]], 1
+ ; CHECK-NEXT: [[Select_FPR64_Using_CC_GPR:%[0-9]+]]:fpr64 = Select_FPR64_Using_CC_GPR [[ANDI]], $x0, 1, [[COPY1]], [[COPY2]]
+ ; CHECK-NEXT: $f10_d = COPY [[Select_FPR64_Using_CC_GPR]]
+ ; CHECK-NEXT: PseudoRET implicit $f10_d
+ %0:gprb(s64) = COPY $x10
+ %1:fprb(s64) = COPY $f10_d
+ %2:fprb(s64) = COPY $f11_d
+ %3:gprb(s64) = G_CONSTANT i64 1
+ %4:gprb(s64) = G_AND %0, %3
+ %5:fprb(s64) = G_SELECT %4(s64), %1, %2
+ $f10_d = COPY %5(s64)
+ PseudoRET implicit $f10_d
+
+...
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-select-rv32.mir b/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-select-rv32.mir
new file mode 100644
index 000000000000000..312fadca866206c
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-select-rv32.mir
@@ -0,0 +1,34 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=riscv32 -mattr=+d -run-pass=regbankselect \
+# RUN: -disable-gisel-legality-check -simplify-mir -verify-machineinstrs %s \
+# RUN: -o - | FileCheck -check-prefix=RV32I %s
+
+---
+name: fp_select_s32
+legalized: true
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $f10_f, $f11_f
+
+ ; RV32I-LABEL: name: fp_select_s32
+ ; RV32I: liveins: $x10, $f10_f, $f11_f
+ ; RV32I-NEXT: {{ $}}
+ ; RV32I-NEXT: [[COPY:%[0-9]+]]:gprb(s32) = COPY $x10
+ ; RV32I-NEXT: [[COPY1:%[0-9]+]]:fprb(s32) = COPY $f10_f
+ ; RV32I-NEXT: [[COPY2:%[0-9]+]]:fprb(s32) = COPY $f11_f
+ ; RV32I-NEXT: [[C:%[0-9]+]]:gprb(s32) = G_CONSTANT i32 1
+ ; RV32I-NEXT: [[AND:%[0-9]+]]:gprb(s32) = G_AND [[COPY]], [[C]]
+ ; RV32I-NEXT: [[SELECT:%[0-9]+]]:fprb(s32) = G_SELECT [[AND]](s32), [[COPY1]], [[COPY2]]
+ ; RV32I-NEXT: $f10_f = COPY [[SELECT]](s32)
+ ; RV32I-NEXT: PseudoRET implicit $f10_f
+ %3:_(s32) = COPY $x10
+ %4:_(s32) = COPY $f10_f
+ %5:_(s32) = COPY $f11_f
+ %12:_(s32) = G_CONSTANT i32 1
+ %11:_(s32) = G_AND %3, %12
+ %10:_(s32) = G_SELECT %11(s32), %4, %5
+ $f10_f = COPY %10(s32)
+ PseudoRET implicit $f10_f
+
+...
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-select-rv64.mir b/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-select-rv64.mir
new file mode 100644
index 000000000000000..22ebd5e93b01407
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-select-rv64.mir
@@ -0,0 +1,34 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=riscv64 -mattr=+d -run-pass=regbankselect \
+# RUN: -disable-gisel-legality-check -simplify-mir -verify-machineinstrs %s \
+# RUN: -o - | FileCheck -check-prefix=RV32I %s
+
+---
+name: fp_select_s64
+legalized: true
+tracksRegLiveness: true
+body: |
+ bb.0.entry:
+ liveins: $x10, $f10_d, $f11_d
+
+ ; RV32I-LABEL: name: fp_select_s64
+ ; RV32I: liveins: $x10, $f10_d, $f11_d
+ ; RV32I-NEXT: {{ $}}
+ ; RV32I-NEXT: [[COPY:%[0-9]+]]:gprb(s64) = COPY $x10
+ ; RV32I-NEXT: [[COPY1:%[0-9]+]]:fprb(s64) = COPY $f10_d
+ ; RV32I-NEXT: [[COPY2:%[0-9]+]]:fprb(s64) = COPY $f11_d
+ ; RV32I-NEXT: [[C:%[0-9]+]]:gprb(s64) = G_CONSTANT i64 1
+ ; RV32I-NEXT: [[AND:%[0-9]+]]:gprb(s64) = G_AND [[COPY]], [[C]]
+ ; RV32I-NEXT: [[SELECT:%[0-9]+]]:fprb(s64) = G_SELECT [[AND]](s64), [[COPY1]], [[COPY2]]
+ ; RV32I-NEXT: $f10_d = COPY [[SELECT]](s64)
+ ; RV32I-NEXT: PseudoRET implicit $f10_d
+ %3:_(s64) = COPY $x10
+ %4:_(s64) = COPY $f10_d
+ %5:_(s64) = COPY $f11_d
+ %12:_(s64) = G_CONSTANT i64 1
+ %11:_(s64) = G_AND %3, %12
+ %10:_(s64) = G_SELECT %11(s64), %4, %5
+ $f10_d = COPY %10(s64)
+ PseudoRET implicit $f10_d
+
+...
``````````
</details>
https://github.com/llvm/llvm-project/pull/72726
More information about the llvm-commits
mailing list