Skip to content

[RISCV][GISel] Use libcalls for f32/f64 G_FCMP without F/D extensions. #117660

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 3 commits into from
Nov 26, 2024
Merged

[RISCV][GISel] Use libcalls for f32/f64 G_FCMP without F/D extensions. #117660

merged 3 commits into from
Nov 26, 2024

Conversation

@topperc
Copy link
Collaborator

@topperc topperc commented Nov 26, 2024

LegalizerHelp only supported f128 libcalls and incorrectly assumed that the destination register for the G_FCMP was s32.

@topperc
[RISCV][GISel] Use libcalls for f32/f64 G_FCMP without F/D extensions.
4a6235b 
LegalizerHelp only supported f128 libcalls and incorrectly assumed
that the destination register for the G_FCMP was s32.
@topperc topperc requested review from aemerson and arsenm November 26, 2024 01:56
@llvmbot
Copy link
Member

llvmbot commented Nov 26, 2024

@llvm/pr-subscribers-llvm-globalisel

Author: Craig Topper (topperc)

Changes

LegalizerHelp only supported f128 libcalls and incorrectly assumed that the destination register for the G_FCMP was s32.

Patch is 42.83 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/117660.diff

4 Files Affected:

  • (modified) llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp (+42-22)
  • (modified) llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp (+2-1)
  • (added) llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll (+584)
  • (added) llvm/test/CodeGen/RISCV/GlobalISel/float-fcmp.ll (+568)
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index 321760ef822bc2..29c7d805ea81e2 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -1001,23 +1001,36 @@ LegalizerHelper::createSetStateLibcall(MachineIRBuilder &MIRBuilder,
 /// the ICMP predicate that should be generated to compare with #0
 /// after the libcall.
 static std::pair<RTLIB::Libcall, CmpInst::Predicate>
-getFCMPLibcallDesc(const CmpInst::Predicate Pred) {
+getFCMPLibcallDesc(const CmpInst::Predicate Pred, unsigned Size) {
+#define RTLIBCASE_CMP(LibcallPrefix, ICmpPred)                                 \
+  do {                                                                         \
+    switch (Size) {                                                            \
+    case 32:                                                                   \
+      return {RTLIB::LibcallPrefix##32, ICmpPred};                             \
+    case 64:                                                                   \
+      return {RTLIB::LibcallPrefix##64, ICmpPred};                             \
+    case 128:                                                                  \
+      return {RTLIB::LibcallPrefix##128, ICmpPred};                            \
+    default:                                                                   \
+      llvm_unreachable("unexpected size");                                     \
+    }                                                                          \
+  } while (0)
 
   switch (Pred) {
   case CmpInst::FCMP_OEQ:
-    return {RTLIB::OEQ_F128, CmpInst::ICMP_EQ};
+    RTLIBCASE_CMP(OEQ_F, CmpInst::ICMP_EQ);
   case CmpInst::FCMP_UNE:
-    return {RTLIB::UNE_F128, CmpInst::ICMP_NE};
+    RTLIBCASE_CMP(UNE_F, CmpInst::ICMP_NE);
   case CmpInst::FCMP_OGE:
-    return {RTLIB::OGE_F128, CmpInst::ICMP_SGE};
+    RTLIBCASE_CMP(OGE_F, CmpInst::ICMP_SGE);
   case CmpInst::FCMP_OLT:
-    return {RTLIB::OLT_F128, CmpInst::ICMP_SLT};
+    RTLIBCASE_CMP(OLT_F, CmpInst::ICMP_SLT);
   case CmpInst::FCMP_OLE:
-    return {RTLIB::OLE_F128, CmpInst::ICMP_SLE};
+    RTLIBCASE_CMP(OLE_F, CmpInst::ICMP_SLE);
   case CmpInst::FCMP_OGT:
-    return {RTLIB::OGT_F128, CmpInst::ICMP_SGT};
+    RTLIBCASE_CMP(OGT_F, CmpInst::ICMP_SGT);
   case CmpInst::FCMP_UNO:
-    return {RTLIB::UO_F128, CmpInst::ICMP_NE};
+    RTLIBCASE_CMP(UO_F, CmpInst::ICMP_NE);
   default:
     return {RTLIB::UNKNOWN_LIBCALL, CmpInst::BAD_ICMP_PREDICATE};
   }
@@ -1032,21 +1045,24 @@ LegalizerHelper::createFCMPLibcall(MachineIRBuilder &MIRBuilder,
   const GFCmp *Cmp = cast<GFCmp>(&MI);
 
   LLT OpLLT = MRI.getType(Cmp->getLHSReg());
-  if (OpLLT != LLT::scalar(128) || OpLLT != MRI.getType(Cmp->getRHSReg()))
+  unsigned Size = OpLLT.getSizeInBits();
+  if ((Size != 32 && Size != 64 && Size != 128) ||
+      OpLLT != MRI.getType(Cmp->getRHSReg()))
     return UnableToLegalize;
 
   Type *OpType = getFloatTypeForLLT(Ctx, OpLLT);
 
   // DstReg type is s32
   const Register DstReg = Cmp->getReg(0);
+  LLT DstTy = MRI.getType(DstReg);
   const auto Cond = Cmp->getCond();
 
   // Reference:
   // https://gcc.gnu.org/onlinedocs/gccint/Soft-float-library-routines.html#Comparison-functions-1
   // Generates a libcall followed by ICMP.
-  const auto BuildLibcall =
-      [&](const RTLIB::Libcall Libcall, const CmpInst::Predicate ICmpPred,
-          const DstOp &Res = LLT::scalar(32)) -> Register {
+  const auto BuildLibcall = [&](const RTLIB::Libcall Libcall,
+                                const CmpInst::Predicate ICmpPred,
+                                const DstOp &Res) -> Register {
     // FCMP libcall always returns an i32, and needs an ICMP with #0.
     constexpr LLT TempLLT = LLT::scalar(32);
     Register Temp = MRI.createGenericVirtualRegister(TempLLT);
@@ -1065,7 +1081,7 @@ LegalizerHelper::createFCMPLibcall(MachineIRBuilder &MIRBuilder,
   };
 
   // Simple case if we have a direct mapping from predicate to libcall
-  if (const auto [Libcall, ICmpPred] = getFCMPLibcallDesc(Cond);
+  if (const auto [Libcall, ICmpPred] = getFCMPLibcallDesc(Cond, Size);
       Libcall != RTLIB::UNKNOWN_LIBCALL &&
       ICmpPred != CmpInst::BAD_ICMP_PREDICATE) {
     if (BuildLibcall(Libcall, ICmpPred, DstReg)) {
@@ -1081,11 +1097,13 @@ LegalizerHelper::createFCMPLibcall(MachineIRBuilder &MIRBuilder,
     // FCMP_UEQ: unordered or equal
     // Convert into (FCMP_OEQ || FCMP_UNO).
 
-    const auto [OeqLibcall, OeqPred] = getFCMPLibcallDesc(CmpInst::FCMP_OEQ);
-    const auto Oeq = BuildLibcall(OeqLibcall, OeqPred);
+    const auto [OeqLibcall, OeqPred] =
+        getFCMPLibcallDesc(CmpInst::FCMP_OEQ, Size);
+    const auto Oeq = BuildLibcall(OeqLibcall, OeqPred, DstTy);
 
-    const auto [UnoLibcall, UnoPred] = getFCMPLibcallDesc(CmpInst::FCMP_UNO);
-    const auto Uno = BuildLibcall(UnoLibcall, UnoPred);
+    const auto [UnoLibcall, UnoPred] =
+        getFCMPLibcallDesc(CmpInst::FCMP_UNO, Size);
+    const auto Uno = BuildLibcall(UnoLibcall, UnoPred, DstTy);
     if (Oeq && Uno)
       MIRBuilder.buildOr(DstReg, Oeq, Uno);
     else
@@ -1100,13 +1118,15 @@ LegalizerHelper::createFCMPLibcall(MachineIRBuilder &MIRBuilder,
     // We inverse the predicate instead of generating a NOT
     // to save one instruction.
     // On AArch64 isel can even select two cmp into a single ccmp.
-    const auto [OeqLibcall, OeqPred] = getFCMPLibcallDesc(CmpInst::FCMP_OEQ);
+    const auto [OeqLibcall, OeqPred] =
+        getFCMPLibcallDesc(CmpInst::FCMP_OEQ, Size);
     const auto NotOeq =
-        BuildLibcall(OeqLibcall, CmpInst::getInversePredicate(OeqPred));
+        BuildLibcall(OeqLibcall, CmpInst::getInversePredicate(OeqPred), DstTy);
 
-    const auto [UnoLibcall, UnoPred] = getFCMPLibcallDesc(CmpInst::FCMP_UNO);
+    const auto [UnoLibcall, UnoPred] =
+        getFCMPLibcallDesc(CmpInst::FCMP_UNO, Size);
     const auto NotUno =
-        BuildLibcall(UnoLibcall, CmpInst::getInversePredicate(UnoPred));
+        BuildLibcall(UnoLibcall, CmpInst::getInversePredicate(UnoPred), DstTy);
 
     if (NotOeq && NotUno)
       MIRBuilder.buildAnd(DstReg, NotOeq, NotUno);
@@ -1128,7 +1148,7 @@ LegalizerHelper::createFCMPLibcall(MachineIRBuilder &MIRBuilder,
     //       MIRBuilder.buildFCmp(CmpInst::getInversePredicate(Pred), PredTy,
     //                            Op1, Op2));
     const auto [InversedLibcall, InversedPred] =
-        getFCMPLibcallDesc(CmpInst::getInversePredicate(Cond));
+        getFCMPLibcallDesc(CmpInst::getInversePredicate(Cond), Size);
     if (!BuildLibcall(InversedLibcall,
                       CmpInst::getInversePredicate(InversedPred), DstReg))
       return UnableToLegalize;
diff --git a/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp b/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
index a9294e76f8763f..e7166fdab8c268 100644
--- a/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
+++ b/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
@@ -530,7 +530,8 @@ RISCVLegalizerInfo::RISCVLegalizerInfo(const RISCVSubtarget &ST)
       .legalFor(ST.hasStdExtF(), {{sXLen, s32}})
       .legalFor(ST.hasStdExtD(), {{sXLen, s64}})
       .legalFor(ST.hasStdExtZfh(), {{sXLen, s16}})
-      .clampScalar(ST.hasStdExtF(), 0, sXLen, sXLen);
+      .clampScalar(0, sXLen, sXLen)
+      .libcallFor({{sXLen, s32}, {sXLen, s64}});
 
   // TODO: Support vector version of G_IS_FPCLASS.
   getActionDefinitionsBuilder(G_IS_FPCLASS)
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll b/llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll
new file mode 100644
index 00000000000000..ce3bd794477302
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll
@@ -0,0 +1,584 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -global-isel -mattr=+d -verify-machineinstrs < %s \
+; RUN:   -target-abi=ilp32d | FileCheck -check-prefix=CHECKIFD %s
+; RUN: llc -mtriple=riscv64 -global-isel -mattr=+d -verify-machineinstrs < %s \
+; RUN:   -target-abi=lp64d | FileCheck -check-prefix=CHECKIFD %s
+; RUN: llc -mtriple=riscv32 -global-isel -verify-machineinstrs < %s \
+; RUN:   | FileCheck -check-prefix=RV32I %s
+; RUN: llc -mtriple=riscv64 -global-isel -verify-machineinstrs < %s \
+; RUN:   | FileCheck -check-prefix=RV64I %s
+
+define i32 @fcmp_false(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_false:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    li a0, 0
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_false:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    li a0, 0
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_false:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    li a0, 0
+; RV64I-NEXT:    ret
+  %1 = fcmp false double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_oeq(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_oeq:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    feq.d a0, fa0, fa1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_oeq:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __eqdf2
+; RV32I-NEXT:    seqz a0, a0
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_oeq:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __eqdf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp oeq double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ogt(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_ogt:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    flt.d a0, fa1, fa0
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_ogt:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __gtdf2
+; RV32I-NEXT:    sgtz a0, a0
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_ogt:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __gtdf2
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    sgtz a0, a0
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp ogt double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_oge(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_oge:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    fle.d a0, fa1, fa0
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_oge:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __gedf2
+; RV32I-NEXT:    slti a0, a0, 0
+; RV32I-NEXT:    xori a0, a0, 1
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_oge:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __gedf2
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    slti a0, a0, 0
+; RV64I-NEXT:    xori a0, a0, 1
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp oge double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_olt(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_olt:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    flt.d a0, fa0, fa1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_olt:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __ltdf2
+; RV32I-NEXT:    slti a0, a0, 0
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_olt:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __ltdf2
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    slti a0, a0, 0
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp olt double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ole(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_ole:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    fle.d a0, fa0, fa1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_ole:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __ledf2
+; RV32I-NEXT:    sgtz a0, a0
+; RV32I-NEXT:    xori a0, a0, 1
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_ole:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __ledf2
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    sgtz a0, a0
+; RV64I-NEXT:    xori a0, a0, 1
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp ole double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_one(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_one:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    flt.d a0, fa0, fa1
+; CHECKIFD-NEXT:    flt.d a1, fa1, fa0
+; CHECKIFD-NEXT:    or a0, a0, a1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_one:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -32
+; RV32I-NEXT:    sw ra, 28(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s0, 24(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s1, 20(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s2, 16(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s3, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s4, 8(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    mv s0, a0
+; RV32I-NEXT:    mv s1, a1
+; RV32I-NEXT:    mv s2, a2
+; RV32I-NEXT:    mv s3, a3
+; RV32I-NEXT:    call __eqdf2
+; RV32I-NEXT:    snez s4, a0
+; RV32I-NEXT:    mv a0, s0
+; RV32I-NEXT:    mv a1, s1
+; RV32I-NEXT:    mv a2, s2
+; RV32I-NEXT:    mv a3, s3
+; RV32I-NEXT:    call __unorddf2
+; RV32I-NEXT:    seqz a0, a0
+; RV32I-NEXT:    and a0, s4, a0
+; RV32I-NEXT:    lw ra, 28(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s0, 24(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s1, 20(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s2, 16(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s3, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s4, 8(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 32
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_one:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -32
+; RV64I-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s0, 16(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s1, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    mv s0, a0
+; RV64I-NEXT:    mv s1, a1
+; RV64I-NEXT:    call __eqdf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    snez s2, a0
+; RV64I-NEXT:    mv a0, s0
+; RV64I-NEXT:    mv a1, s1
+; RV64I-NEXT:    call __unorddf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    and a0, s2, a0
+; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s2, 0(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 32
+; RV64I-NEXT:    ret
+  %1 = fcmp one double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ord(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_ord:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    feq.d a0, fa0, fa0
+; CHECKIFD-NEXT:    feq.d a1, fa1, fa1
+; CHECKIFD-NEXT:    and a0, a0, a1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_ord:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __unorddf2
+; RV32I-NEXT:    seqz a0, a0
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_ord:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __unorddf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp ord double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ueq(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_ueq:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    flt.d a0, fa0, fa1
+; CHECKIFD-NEXT:    flt.d a1, fa1, fa0
+; CHECKIFD-NEXT:    or a0, a0, a1
+; CHECKIFD-NEXT:    xori a0, a0, 1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_ueq:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -32
+; RV32I-NEXT:    sw ra, 28(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s0, 24(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s1, 20(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s2, 16(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s3, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s4, 8(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    mv s0, a0
+; RV32I-NEXT:    mv s1, a1
+; RV32I-NEXT:    mv s2, a2
+; RV32I-NEXT:    mv s3, a3
+; RV32I-NEXT:    call __eqdf2
+; RV32I-NEXT:    seqz s4, a0
+; RV32I-NEXT:    mv a0, s0
+; RV32I-NEXT:    mv a1, s1
+; RV32I-NEXT:    mv a2, s2
+; RV32I-NEXT:    mv a3, s3
+; RV32I-NEXT:    call __unorddf2
+; RV32I-NEXT:    snez a0, a0
+; RV32I-NEXT:    or a0, s4, a0
+; RV32I-NEXT:    lw ra, 28(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s0, 24(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s1, 20(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s2, 16(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s3, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s4, 8(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 32
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_ueq:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -32
+; RV64I-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s0, 16(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s1, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    mv s0, a0
+; RV64I-NEXT:    mv s1, a1
+; RV64I-NEXT:    call __eqdf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    seqz s2, a0
+; RV64I-NEXT:    mv a0, s0
+; RV64I-NEXT:    mv a1, s1
+; RV64I-NEXT:    call __unorddf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    snez a0, a0
+; RV64I-NEXT:    or a0, s2, a0
+; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s2, 0(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 32
+; RV64I-NEXT:    ret
+  %1 = fcmp ueq double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ugt(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_ugt:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    fle.d a0, fa0, fa1
+; CHECKIFD-NEXT:    xori a0, a0, 1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_ugt:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __ledf2
+; RV32I-NEXT:    sgtz a0, a0
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_ugt:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __ledf2
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    sgtz a0, a0
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-N...
[truncated]

@llvmbot
Copy link
Member

llvmbot commented Nov 26, 2024

@llvm/pr-subscribers-backend-risc-v

Author: Craig Topper (topperc)

Changes

LegalizerHelp only supported f128 libcalls and incorrectly assumed that the destination register for the G_FCMP was s32.

Patch is 42.83 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/117660.diff

4 Files Affected:

  • (modified) llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp (+42-22)
  • (modified) llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp (+2-1)
  • (added) llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll (+584)
  • (added) llvm/test/CodeGen/RISCV/GlobalISel/float-fcmp.ll (+568)
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index 321760ef822bc2..29c7d805ea81e2 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -1001,23 +1001,36 @@ LegalizerHelper::createSetStateLibcall(MachineIRBuilder &MIRBuilder,
 /// the ICMP predicate that should be generated to compare with #0
 /// after the libcall.
 static std::pair<RTLIB::Libcall, CmpInst::Predicate>
-getFCMPLibcallDesc(const CmpInst::Predicate Pred) {
+getFCMPLibcallDesc(const CmpInst::Predicate Pred, unsigned Size) {
+#define RTLIBCASE_CMP(LibcallPrefix, ICmpPred)                                 \
+  do {                                                                         \
+    switch (Size) {                                                            \
+    case 32:                                                                   \
+      return {RTLIB::LibcallPrefix##32, ICmpPred};                             \
+    case 64:                                                                   \
+      return {RTLIB::LibcallPrefix##64, ICmpPred};                             \
+    case 128:                                                                  \
+      return {RTLIB::LibcallPrefix##128, ICmpPred};                            \
+    default:                                                                   \
+      llvm_unreachable("unexpected size");                                     \
+    }                                                                          \
+  } while (0)
 
   switch (Pred) {
   case CmpInst::FCMP_OEQ:
-    return {RTLIB::OEQ_F128, CmpInst::ICMP_EQ};
+    RTLIBCASE_CMP(OEQ_F, CmpInst::ICMP_EQ);
   case CmpInst::FCMP_UNE:
-    return {RTLIB::UNE_F128, CmpInst::ICMP_NE};
+    RTLIBCASE_CMP(UNE_F, CmpInst::ICMP_NE);
   case CmpInst::FCMP_OGE:
-    return {RTLIB::OGE_F128, CmpInst::ICMP_SGE};
+    RTLIBCASE_CMP(OGE_F, CmpInst::ICMP_SGE);
   case CmpInst::FCMP_OLT:
-    return {RTLIB::OLT_F128, CmpInst::ICMP_SLT};
+    RTLIBCASE_CMP(OLT_F, CmpInst::ICMP_SLT);
   case CmpInst::FCMP_OLE:
-    return {RTLIB::OLE_F128, CmpInst::ICMP_SLE};
+    RTLIBCASE_CMP(OLE_F, CmpInst::ICMP_SLE);
   case CmpInst::FCMP_OGT:
-    return {RTLIB::OGT_F128, CmpInst::ICMP_SGT};
+    RTLIBCASE_CMP(OGT_F, CmpInst::ICMP_SGT);
   case CmpInst::FCMP_UNO:
-    return {RTLIB::UO_F128, CmpInst::ICMP_NE};
+    RTLIBCASE_CMP(UO_F, CmpInst::ICMP_NE);
   default:
     return {RTLIB::UNKNOWN_LIBCALL, CmpInst::BAD_ICMP_PREDICATE};
   }
@@ -1032,21 +1045,24 @@ LegalizerHelper::createFCMPLibcall(MachineIRBuilder &MIRBuilder,
   const GFCmp *Cmp = cast<GFCmp>(&MI);
 
   LLT OpLLT = MRI.getType(Cmp->getLHSReg());
-  if (OpLLT != LLT::scalar(128) || OpLLT != MRI.getType(Cmp->getRHSReg()))
+  unsigned Size = OpLLT.getSizeInBits();
+  if ((Size != 32 && Size != 64 && Size != 128) ||
+      OpLLT != MRI.getType(Cmp->getRHSReg()))
     return UnableToLegalize;
 
   Type *OpType = getFloatTypeForLLT(Ctx, OpLLT);
 
   // DstReg type is s32
   const Register DstReg = Cmp->getReg(0);
+  LLT DstTy = MRI.getType(DstReg);
   const auto Cond = Cmp->getCond();
 
   // Reference:
   // https://gcc.gnu.org/onlinedocs/gccint/Soft-float-library-routines.html#Comparison-functions-1
   // Generates a libcall followed by ICMP.
-  const auto BuildLibcall =
-      [&](const RTLIB::Libcall Libcall, const CmpInst::Predicate ICmpPred,
-          const DstOp &Res = LLT::scalar(32)) -> Register {
+  const auto BuildLibcall = [&](const RTLIB::Libcall Libcall,
+                                const CmpInst::Predicate ICmpPred,
+                                const DstOp &Res) -> Register {
     // FCMP libcall always returns an i32, and needs an ICMP with #0.
     constexpr LLT TempLLT = LLT::scalar(32);
     Register Temp = MRI.createGenericVirtualRegister(TempLLT);
@@ -1065,7 +1081,7 @@ LegalizerHelper::createFCMPLibcall(MachineIRBuilder &MIRBuilder,
   };
 
   // Simple case if we have a direct mapping from predicate to libcall
-  if (const auto [Libcall, ICmpPred] = getFCMPLibcallDesc(Cond);
+  if (const auto [Libcall, ICmpPred] = getFCMPLibcallDesc(Cond, Size);
       Libcall != RTLIB::UNKNOWN_LIBCALL &&
       ICmpPred != CmpInst::BAD_ICMP_PREDICATE) {
     if (BuildLibcall(Libcall, ICmpPred, DstReg)) {
@@ -1081,11 +1097,13 @@ LegalizerHelper::createFCMPLibcall(MachineIRBuilder &MIRBuilder,
     // FCMP_UEQ: unordered or equal
     // Convert into (FCMP_OEQ || FCMP_UNO).
 
-    const auto [OeqLibcall, OeqPred] = getFCMPLibcallDesc(CmpInst::FCMP_OEQ);
-    const auto Oeq = BuildLibcall(OeqLibcall, OeqPred);
+    const auto [OeqLibcall, OeqPred] =
+        getFCMPLibcallDesc(CmpInst::FCMP_OEQ, Size);
+    const auto Oeq = BuildLibcall(OeqLibcall, OeqPred, DstTy);
 
-    const auto [UnoLibcall, UnoPred] = getFCMPLibcallDesc(CmpInst::FCMP_UNO);
-    const auto Uno = BuildLibcall(UnoLibcall, UnoPred);
+    const auto [UnoLibcall, UnoPred] =
+        getFCMPLibcallDesc(CmpInst::FCMP_UNO, Size);
+    const auto Uno = BuildLibcall(UnoLibcall, UnoPred, DstTy);
     if (Oeq && Uno)
       MIRBuilder.buildOr(DstReg, Oeq, Uno);
     else
@@ -1100,13 +1118,15 @@ LegalizerHelper::createFCMPLibcall(MachineIRBuilder &MIRBuilder,
     // We inverse the predicate instead of generating a NOT
     // to save one instruction.
     // On AArch64 isel can even select two cmp into a single ccmp.
-    const auto [OeqLibcall, OeqPred] = getFCMPLibcallDesc(CmpInst::FCMP_OEQ);
+    const auto [OeqLibcall, OeqPred] =
+        getFCMPLibcallDesc(CmpInst::FCMP_OEQ, Size);
     const auto NotOeq =
-        BuildLibcall(OeqLibcall, CmpInst::getInversePredicate(OeqPred));
+        BuildLibcall(OeqLibcall, CmpInst::getInversePredicate(OeqPred), DstTy);
 
-    const auto [UnoLibcall, UnoPred] = getFCMPLibcallDesc(CmpInst::FCMP_UNO);
+    const auto [UnoLibcall, UnoPred] =
+        getFCMPLibcallDesc(CmpInst::FCMP_UNO, Size);
     const auto NotUno =
-        BuildLibcall(UnoLibcall, CmpInst::getInversePredicate(UnoPred));
+        BuildLibcall(UnoLibcall, CmpInst::getInversePredicate(UnoPred), DstTy);
 
     if (NotOeq && NotUno)
       MIRBuilder.buildAnd(DstReg, NotOeq, NotUno);
@@ -1128,7 +1148,7 @@ LegalizerHelper::createFCMPLibcall(MachineIRBuilder &MIRBuilder,
     //       MIRBuilder.buildFCmp(CmpInst::getInversePredicate(Pred), PredTy,
     //                            Op1, Op2));
     const auto [InversedLibcall, InversedPred] =
-        getFCMPLibcallDesc(CmpInst::getInversePredicate(Cond));
+        getFCMPLibcallDesc(CmpInst::getInversePredicate(Cond), Size);
     if (!BuildLibcall(InversedLibcall,
                       CmpInst::getInversePredicate(InversedPred), DstReg))
       return UnableToLegalize;
diff --git a/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp b/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
index a9294e76f8763f..e7166fdab8c268 100644
--- a/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
+++ b/llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp
@@ -530,7 +530,8 @@ RISCVLegalizerInfo::RISCVLegalizerInfo(const RISCVSubtarget &ST)
       .legalFor(ST.hasStdExtF(), {{sXLen, s32}})
       .legalFor(ST.hasStdExtD(), {{sXLen, s64}})
       .legalFor(ST.hasStdExtZfh(), {{sXLen, s16}})
-      .clampScalar(ST.hasStdExtF(), 0, sXLen, sXLen);
+      .clampScalar(0, sXLen, sXLen)
+      .libcallFor({{sXLen, s32}, {sXLen, s64}});
 
   // TODO: Support vector version of G_IS_FPCLASS.
   getActionDefinitionsBuilder(G_IS_FPCLASS)
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll b/llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll
new file mode 100644
index 00000000000000..ce3bd794477302
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll
@@ -0,0 +1,584 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -global-isel -mattr=+d -verify-machineinstrs < %s \
+; RUN:   -target-abi=ilp32d | FileCheck -check-prefix=CHECKIFD %s
+; RUN: llc -mtriple=riscv64 -global-isel -mattr=+d -verify-machineinstrs < %s \
+; RUN:   -target-abi=lp64d | FileCheck -check-prefix=CHECKIFD %s
+; RUN: llc -mtriple=riscv32 -global-isel -verify-machineinstrs < %s \
+; RUN:   | FileCheck -check-prefix=RV32I %s
+; RUN: llc -mtriple=riscv64 -global-isel -verify-machineinstrs < %s \
+; RUN:   | FileCheck -check-prefix=RV64I %s
+
+define i32 @fcmp_false(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_false:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    li a0, 0
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_false:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    li a0, 0
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_false:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    li a0, 0
+; RV64I-NEXT:    ret
+  %1 = fcmp false double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_oeq(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_oeq:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    feq.d a0, fa0, fa1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_oeq:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __eqdf2
+; RV32I-NEXT:    seqz a0, a0
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_oeq:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __eqdf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp oeq double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ogt(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_ogt:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    flt.d a0, fa1, fa0
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_ogt:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __gtdf2
+; RV32I-NEXT:    sgtz a0, a0
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_ogt:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __gtdf2
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    sgtz a0, a0
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp ogt double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_oge(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_oge:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    fle.d a0, fa1, fa0
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_oge:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __gedf2
+; RV32I-NEXT:    slti a0, a0, 0
+; RV32I-NEXT:    xori a0, a0, 1
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_oge:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __gedf2
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    slti a0, a0, 0
+; RV64I-NEXT:    xori a0, a0, 1
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp oge double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_olt(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_olt:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    flt.d a0, fa0, fa1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_olt:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __ltdf2
+; RV32I-NEXT:    slti a0, a0, 0
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_olt:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __ltdf2
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    slti a0, a0, 0
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp olt double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ole(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_ole:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    fle.d a0, fa0, fa1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_ole:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __ledf2
+; RV32I-NEXT:    sgtz a0, a0
+; RV32I-NEXT:    xori a0, a0, 1
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_ole:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __ledf2
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    sgtz a0, a0
+; RV64I-NEXT:    xori a0, a0, 1
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp ole double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_one(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_one:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    flt.d a0, fa0, fa1
+; CHECKIFD-NEXT:    flt.d a1, fa1, fa0
+; CHECKIFD-NEXT:    or a0, a0, a1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_one:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -32
+; RV32I-NEXT:    sw ra, 28(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s0, 24(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s1, 20(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s2, 16(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s3, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s4, 8(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    mv s0, a0
+; RV32I-NEXT:    mv s1, a1
+; RV32I-NEXT:    mv s2, a2
+; RV32I-NEXT:    mv s3, a3
+; RV32I-NEXT:    call __eqdf2
+; RV32I-NEXT:    snez s4, a0
+; RV32I-NEXT:    mv a0, s0
+; RV32I-NEXT:    mv a1, s1
+; RV32I-NEXT:    mv a2, s2
+; RV32I-NEXT:    mv a3, s3
+; RV32I-NEXT:    call __unorddf2
+; RV32I-NEXT:    seqz a0, a0
+; RV32I-NEXT:    and a0, s4, a0
+; RV32I-NEXT:    lw ra, 28(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s0, 24(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s1, 20(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s2, 16(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s3, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s4, 8(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 32
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_one:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -32
+; RV64I-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s0, 16(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s1, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    mv s0, a0
+; RV64I-NEXT:    mv s1, a1
+; RV64I-NEXT:    call __eqdf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    snez s2, a0
+; RV64I-NEXT:    mv a0, s0
+; RV64I-NEXT:    mv a1, s1
+; RV64I-NEXT:    call __unorddf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    and a0, s2, a0
+; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s2, 0(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 32
+; RV64I-NEXT:    ret
+  %1 = fcmp one double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ord(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_ord:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    feq.d a0, fa0, fa0
+; CHECKIFD-NEXT:    feq.d a1, fa1, fa1
+; CHECKIFD-NEXT:    and a0, a0, a1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_ord:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __unorddf2
+; RV32I-NEXT:    seqz a0, a0
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_ord:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __unorddf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 16
+; RV64I-NEXT:    ret
+  %1 = fcmp ord double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ueq(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_ueq:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    flt.d a0, fa0, fa1
+; CHECKIFD-NEXT:    flt.d a1, fa1, fa0
+; CHECKIFD-NEXT:    or a0, a0, a1
+; CHECKIFD-NEXT:    xori a0, a0, 1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_ueq:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -32
+; RV32I-NEXT:    sw ra, 28(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s0, 24(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s1, 20(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s2, 16(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s3, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    sw s4, 8(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    mv s0, a0
+; RV32I-NEXT:    mv s1, a1
+; RV32I-NEXT:    mv s2, a2
+; RV32I-NEXT:    mv s3, a3
+; RV32I-NEXT:    call __eqdf2
+; RV32I-NEXT:    seqz s4, a0
+; RV32I-NEXT:    mv a0, s0
+; RV32I-NEXT:    mv a1, s1
+; RV32I-NEXT:    mv a2, s2
+; RV32I-NEXT:    mv a3, s3
+; RV32I-NEXT:    call __unorddf2
+; RV32I-NEXT:    snez a0, a0
+; RV32I-NEXT:    or a0, s4, a0
+; RV32I-NEXT:    lw ra, 28(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s0, 24(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s1, 20(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s2, 16(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s3, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    lw s4, 8(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 32
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_ueq:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -32
+; RV64I-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s0, 16(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s1, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    mv s0, a0
+; RV64I-NEXT:    mv s1, a1
+; RV64I-NEXT:    call __eqdf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    seqz s2, a0
+; RV64I-NEXT:    mv a0, s0
+; RV64I-NEXT:    mv a1, s1
+; RV64I-NEXT:    call __unorddf2
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    snez a0, a0
+; RV64I-NEXT:    or a0, s2, a0
+; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    ld s2, 0(sp) # 8-byte Folded Reload
+; RV64I-NEXT:    addi sp, sp, 32
+; RV64I-NEXT:    ret
+  %1 = fcmp ueq double %a, %b
+  %2 = zext i1 %1 to i32
+  ret i32 %2
+}
+
+define i32 @fcmp_ugt(double %a, double %b) nounwind {
+; CHECKIFD-LABEL: fcmp_ugt:
+; CHECKIFD:       # %bb.0:
+; CHECKIFD-NEXT:    fle.d a0, fa0, fa1
+; CHECKIFD-NEXT:    xori a0, a0, 1
+; CHECKIFD-NEXT:    ret
+;
+; RV32I-LABEL: fcmp_ugt:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi sp, sp, -16
+; RV32I-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT:    call __ledf2
+; RV32I-NEXT:    sgtz a0, a0
+; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT:    addi sp, sp, 16
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: fcmp_ugt:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    addi sp, sp, -16
+; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT:    call __ledf2
+; RV64I-NEXT:    sext.w a0, a0
+; RV64I-NEXT:    sgtz a0, a0
+; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-N...
[truncated]

topperc
topperc commented Nov 26, 2024
View reviewed changes
llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll
; RV64I-NEXT: addi sp, sp, -16
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
; RV64I-NEXT: call __eqdf2
; RV64I-NEXT: slli a0, a0, 32
Copy link
Collaborator Author

@topperc topperc Nov 26, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Still need to figure out how to get rid of this shift pair.

topperc
topperc commented Nov 26, 2024
View reviewed changes
llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll
; RV64I-NEXT: addi sp, sp, -16
; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
; RV64I-NEXT: call __gtdf2
; RV64I-NEXT: sext.w a0, a0
Copy link
Collaborator Author

@topperc topperc Nov 26, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Need to figure out how to get rid of this sext.w

arsenm
arsenm approved these changes Nov 26, 2024
View reviewed changes
llvm/test/CodeGen/RISCV/GlobalISel/double-fcmp.ll
; RV64I: # %bb.0:
; RV64I-NEXT: li a0, 0
; RV64I-NEXT: ret
%1 = fcmp false double %a, %b
Copy link
Contributor

@arsenm arsenm Nov 26, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Named values

Copy link
Collaborator Author

@topperc topperc Nov 26, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This was copied from an existing SelectionDAG test. I'd prefer to keep them in sync for easier comparison.

Copy link
Contributor

@arsenm arsenm Nov 26, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should just merge the tests with multiple run lines then?

Copy link
Collaborator Author

@topperc topperc Nov 26, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Eventually I probably will. This is the first FP IR test I was able to completely copy and get it to pass. I've copied other tests, but had to delete an individual test case for one reason or another that I'll need to come back to.

@topperc topperc merged commit 43b6b78 into llvm:main Nov 26, 2024
5 of 7 checks passed
@topperc topperc deleted the pr/fcmp branch November 26, 2024 23:48
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@arsenm arsenm arsenm approved these changes

@aemerson aemerson Awaiting requested review from aemerson

Assignees

No one assigned

Labels

backend:RISC-V llvm:globalisel

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@topperc @llvmbot @arsenm