[X86][GlobalISel] Added support for llvm.set.rounding #156591
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Contributor
JaydeepChauhan14
commented
Sep 3, 2025
JaydeepChauhan14
commented
- This implementation is adapted from SDAG X86TargetLowering::LowerSET_ROUNDING.
Contributor
Author
Member
|
@llvm/pr-subscribers-backend-risc-v @llvm/pr-subscribers-llvm-globalisel Author: None (JaydeepChauhan14) Changes
Patch is 32.85 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/156591.diff 12 Files Affected:
diff --git a/llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h b/llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h
index 99d3cd0aac85c..0b6033b4ba60a 100644
--- a/llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h
@@ -2463,6 +2463,11 @@ class LLVM_ABI MachineIRBuilder {
return buildInstr(TargetOpcode::G_GET_ROUNDING, {Dst}, {});
}
+ /// Build and insert G_SET_ROUNDING
+ MachineInstrBuilder buildSetRounding(const SrcOp &Src) {
+ return buildInstr(TargetOpcode::G_SET_ROUNDING, {}, {Src});
+ }
+
virtual MachineInstrBuilder
buildInstr(unsigned Opc, ArrayRef<DstOp> DstOps, ArrayRef<SrcOp> SrcOps,
std::optional<unsigned> Flags = std::nullopt);
diff --git a/llvm/include/llvm/Support/TargetOpcodes.def b/llvm/include/llvm/Support/TargetOpcodes.def
index b905576b61791..7710e2fc2f22b 100644
--- a/llvm/include/llvm/Support/TargetOpcodes.def
+++ b/llvm/include/llvm/Support/TargetOpcodes.def
@@ -745,6 +745,7 @@ HANDLE_TARGET_OPCODE(G_SET_FPMODE)
HANDLE_TARGET_OPCODE(G_RESET_FPMODE)
HANDLE_TARGET_OPCODE(G_GET_ROUNDING)
+HANDLE_TARGET_OPCODE(G_SET_ROUNDING)
/// Generic pointer offset
HANDLE_TARGET_OPCODE(G_PTR_ADD)
diff --git a/llvm/include/llvm/Target/GenericOpcodes.td b/llvm/include/llvm/Target/GenericOpcodes.td
index ce4750db88c9a..733d10b1c5f3c 100644
--- a/llvm/include/llvm/Target/GenericOpcodes.td
+++ b/llvm/include/llvm/Target/GenericOpcodes.td
@@ -1273,6 +1273,12 @@ def G_GET_ROUNDING : GenericInstruction {
let hasSideEffects = true;
}
+def G_SET_ROUNDING : GenericInstruction {
+ let OutOperandList = (outs);
+ let InOperandList = (ins type0:$src);
+ let hasSideEffects = true;
+}
+
//------------------------------------------------------------------------------
// Memory ops
//------------------------------------------------------------------------------
diff --git a/llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp b/llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp
index 541269ab6bfce..dbf482dba7c44 100644
--- a/llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp
@@ -2607,6 +2607,9 @@ bool IRTranslator::translateKnownIntrinsic(const CallInst &CI, Intrinsic::ID ID,
case Intrinsic::get_rounding:
MIRBuilder.buildGetRounding(getOrCreateVReg(CI));
return true;
+ case Intrinsic::set_rounding:
+ MIRBuilder.buildSetRounding(getOrCreateVReg(*CI.getOperand(0)));
+ return true;
case Intrinsic::vscale: {
MIRBuilder.buildVScale(getOrCreateVReg(CI), 1);
return true;
diff --git a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp
index 7fe58539cd4ec..6a63cb327b783 100644
--- a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp
+++ b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp
@@ -21,6 +21,7 @@
#include "llvm/CodeGen/TargetOpcodes.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/IntrinsicsX86.h"
#include "llvm/IR/Type.h"
using namespace llvm;
@@ -109,7 +110,8 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
.legalFor(HasSSE2 || UseX87, {s64})
.legalFor(UseX87, {s80});
- getActionDefinitionsBuilder(G_GET_ROUNDING).customFor({s32});
+ getActionDefinitionsBuilder({G_GET_ROUNDING, G_SET_ROUNDING})
+ .customFor({s32});
// merge/unmerge
for (unsigned Op : {G_MERGE_VALUES, G_UNMERGE_VALUES}) {
@@ -616,6 +618,8 @@ bool X86LegalizerInfo::legalizeCustom(LegalizerHelper &Helper, MachineInstr &MI,
return legalizeFPTOSI(MI, MRI, Helper);
case TargetOpcode::G_GET_ROUNDING:
return legalizeGETROUNDING(MI, MRI, Helper);
+ case TargetOpcode::G_SET_ROUNDING:
+ return legalizeSETROUNDING(MI, MRI, Helper);
}
llvm_unreachable("expected switch to return");
}
@@ -858,6 +862,116 @@ bool X86LegalizerInfo::legalizeGETROUNDING(MachineInstr &MI,
return true;
}
+bool X86LegalizerInfo::legalizeSETROUNDING(MachineInstr &MI,
+ MachineRegisterInfo &MRI,
+ LegalizerHelper &Helper) const {
+ MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
+ MachineFunction &MF = MIRBuilder.getMF();
+ Register Src = MI.getOperand(0).getReg();
+ const LLT s8 = LLT::scalar(8);
+ const LLT s16 = LLT::scalar(16);
+ const LLT s32 = LLT::scalar(32);
+
+ // Allocate stack slot for control word and MXCSR (4 bytes).
+ int MemSize = 4;
+ Align Alignment = Align(4);
+ MachinePointerInfo PtrInfo;
+ auto StackTemp = Helper.createStackTemporary(TypeSize::getFixed(MemSize),
+ Alignment, PtrInfo);
+ Register StackPtr = StackTemp.getReg(0);
+
+ auto StoreMMO =
+ MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore, 2, Align(2));
+ MIRBuilder.buildInstr(X86::G_FNSTCW16)
+ .addUse(StackPtr)
+ .addMemOperand(StoreMMO);
+
+ auto LoadMMO =
+ MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOLoad, 2, Align(2));
+ auto CWD16 = MIRBuilder.buildLoad(s16, StackPtr, *LoadMMO);
+
+ // Clear RM field (bits 11:10)
+ auto ClearedCWD =
+ MIRBuilder.buildAnd(s16, CWD16, MIRBuilder.buildConstant(s16, 0xf3ff));
+
+ // Convert Src (rounding mode) to bits for control word
+ // (0xc9 << (2 * Src + 4)) & 0xc00
+ LLT SrcTy = MRI.getType(Src);
+ Register Src32;
+ if (SrcTy.getSizeInBits() < 32)
+ Src32 = MIRBuilder.buildZExt(s32, Src).getReg(0);
+ else if (SrcTy.getSizeInBits() > 32)
+ Src32 = MIRBuilder.buildTrunc(s32, Src).getReg(0);
+ else
+ Src32 = Src;
+ auto ShiftAmt = MIRBuilder.buildAdd(
+ s32, MIRBuilder.buildShl(s32, Src32, MIRBuilder.buildConstant(s32, 1)),
+ MIRBuilder.buildConstant(s32, 4));
+ auto ShiftAmt8 = MIRBuilder.buildTrunc(s8, ShiftAmt);
+ auto Shifted =
+ MIRBuilder.buildShl(s16, MIRBuilder.buildConstant(s16, 0xc9), ShiftAmt8);
+ auto RMBits =
+ MIRBuilder.buildAnd(s16, Shifted, MIRBuilder.buildConstant(s16, 0xc00));
+
+ // Update rounding mode bits
+ auto NewCWD = MIRBuilder.buildOr(s16, ClearedCWD, RMBits);
+
+ // Store new FP Control Word to stack
+ auto StoreNewMMO =
+ MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore, 2, Align(2));
+ MIRBuilder.buildStore(NewCWD, StackPtr, *StoreNewMMO);
+
+ // Load FP control word from the slot using G_FLDCW16
+ auto LoadNewMMO =
+ MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOLoad, 2, Align(2));
+ MIRBuilder.buildInstr(X86::G_FLDCW16)
+ .addUse(StackPtr)
+ .addMemOperand(LoadNewMMO);
+
+ if (Subtarget.hasSSE1()) {
+ // Store MXCSR to stack (use STMXCSR)
+ auto StoreMXCSRMMO = MF.getMachineMemOperand(
+ PtrInfo, MachineMemOperand::MOStore, 4, Align(4));
+ MIRBuilder.buildInstr(TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS)
+ .addIntrinsicID(Intrinsic::x86_sse_stmxcsr)
+ .addUse(StackPtr)
+ .addMemOperand(StoreMXCSRMMO);
+
+ // Load MXCSR from stack
+ auto LoadMXCSRMMO = MF.getMachineMemOperand(
+ PtrInfo, MachineMemOperand::MOLoad, 4, Align(4));
+ auto MXCSR = MIRBuilder.buildLoad(s32, StackPtr, *LoadMXCSRMMO);
+
+ // Clear RM field (bits 14:13)
+ auto ClearedMXCSR = MIRBuilder.buildAnd(
+ s32, MXCSR, MIRBuilder.buildConstant(s32, 0xffff9fff));
+
+ // Shift x87 RM bits from 11:10 to 14:13
+ auto RMBits32 = MIRBuilder.buildZExt(s32, RMBits);
+ auto MXCSRRMBits =
+ MIRBuilder.buildShl(s32, RMBits32, MIRBuilder.buildConstant(s32, 3));
+
+ // Update rounding mode bits
+ auto NewMXCSR = MIRBuilder.buildOr(s32, ClearedMXCSR, MXCSRRMBits);
+
+ // Store new MXCSR to stack
+ auto StoreNewMXCSRMMO = MF.getMachineMemOperand(
+ PtrInfo, MachineMemOperand::MOStore, 4, Align(4));
+ MIRBuilder.buildStore(NewMXCSR, StackPtr, *StoreNewMXCSRMMO);
+
+ // Load MXCSR from stack (use LDMXCSR)
+ auto LoadNewMXCSRMMO = MF.getMachineMemOperand(
+ PtrInfo, MachineMemOperand::MOLoad, 4, Align(4));
+ MIRBuilder.buildInstr(TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS)
+ .addIntrinsicID(Intrinsic::x86_sse_ldmxcsr)
+ .addUse(StackPtr)
+ .addMemOperand(LoadNewMXCSRMMO);
+ }
+
+ MI.eraseFromParent();
+ return true;
+}
+
bool X86LegalizerInfo::legalizeIntrinsic(LegalizerHelper &Helper,
MachineInstr &MI) const {
return true;
diff --git a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h
index 0003552d70ee0..09c727c8e8685 100644
--- a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h
+++ b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h
@@ -57,6 +57,9 @@ class X86LegalizerInfo : public LegalizerInfo {
bool legalizeGETROUNDING(MachineInstr &MI, MachineRegisterInfo &MRI,
LegalizerHelper &Helper) const;
+
+ bool legalizeSETROUNDING(MachineInstr &MI, MachineRegisterInfo &MRI,
+ LegalizerHelper &Helper) const;
};
} // namespace llvm
#endif
diff --git a/llvm/lib/Target/X86/X86InstrGISel.td b/llvm/lib/Target/X86/X86InstrGISel.td
index 39198214037a3..b0c6bb6f61ad8 100644
--- a/llvm/lib/Target/X86/X86InstrGISel.td
+++ b/llvm/lib/Target/X86/X86InstrGISel.td
@@ -34,6 +34,14 @@ def G_FNSTCW16 : X86GenericInstruction {
let mayStore = true;
}
+def G_FLDCW16 : X86GenericInstruction {
+ let OutOperandList = (outs);
+ let InOperandList = (ins ptype0:$src);
+ let hasSideEffects = true;
+ let mayLoad = true;
+}
+
def : GINodeEquiv<G_FILD, X86fild>;
def : GINodeEquiv<G_FIST, X86fp_to_mem>;
def : GINodeEquiv<G_FNSTCW16, X86fp_cwd_get16>;
+def : GINodeEquiv<G_FLDCW16, X86fp_cwd_set16>;
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
index 040f97f96ee21..9ad28a5487233 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
@@ -642,6 +642,9 @@
# DEBUG-NEXT: G_GET_ROUNDING (opcode {{[0-9]+}}): 1 type index, 0 imm indices
# DEBUG-NEXT:.. type index coverage check SKIPPED: no rules defined
# DEBUG-NEXT:.. imm index coverage check SKIPPED: no rules defined
+# DEBUG-NEXT: G_SET_ROUNDING (opcode {{[0-9]+}}): 1 type index, 0 imm indices
+# DEBUG-NEXT:.. type index coverage check SKIPPED: no rules defined
+# DEBUG-NEXT:.. imm index coverage check SKIPPED: no rules defined
# DEBUG-NEXT: G_PTR_ADD (opcode {{[0-9]+}}): 2 type indices, 0 imm indices
# DEBUG-NEXT: .. the first uncovered type index: 2, OK
# DEBUG-NEXT: .. the first uncovered imm index: 0, OK
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer-info-validation.mir b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer-info-validation.mir
index 9d68a6d72c486..f56e78a645d50 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer-info-validation.mir
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer-info-validation.mir
@@ -629,6 +629,9 @@
# DEBUG-NEXT: G_GET_ROUNDING (opcode {{[0-9]+}}): 1 type index, 0 imm indices
# DEBUG-NEXT:.. type index coverage check SKIPPED: no rules defined
# DEBUG-NEXT:.. imm index coverage check SKIPPED: no rules defined
+# DEBUG-NEXT: G_SET_ROUNDING (opcode {{[0-9]+}}): 1 type index, 0 imm indices
+# DEBUG-NEXT:.. type index coverage check SKIPPED: no rules defined
+# DEBUG-NEXT:.. imm index coverage check SKIPPED: no rules defined
# DEBUG-NEXT: G_PTR_ADD (opcode {{[0-9]+}}): 2 type indices, 0 imm indices
# DEBUG-NEXT: .. the first uncovered type index: 2, OK
# DEBUG-NEXT: .. the first uncovered imm index: 0, OK
diff --git a/llvm/test/CodeGen/X86/isel-llvm.set.rounding.ll b/llvm/test/CodeGen/X86/isel-llvm.set.rounding.ll
index 688add1e92ab1..f589c16ff18da 100644
--- a/llvm/test/CodeGen/X86/isel-llvm.set.rounding.ll
+++ b/llvm/test/CodeGen/X86/isel-llvm.set.rounding.ll
@@ -1,16 +1,16 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i686-- -mattr=-sse | FileCheck %s --check-prefixes=X86-NOSSE,SDAG-X86-NOSSE
; RUN: llc < %s -mtriple=i686-- -fast-isel -fast-isel-abort=1 -mattr=-sse | FileCheck %s --check-prefixes=X86-NOSSE,FASTISEL-X86-NOSSE
-; RUN: llc < %s -mtriple=i686-- -global-isel -global-isel-abort=2 -mattr=-sse | FileCheck %s --check-prefixes=X86-NOSSE,GISEL-X86-NOSSE
+; RUN: llc < %s -mtriple=i686-- -global-isel -global-isel-abort=1 -mattr=-sse | FileCheck %s --check-prefixes=GISEL-X86-NOSSE
; RUN: llc < %s -mtriple=x86_64-- -mattr=-sse | FileCheck %s --check-prefixes=X64-NOSSE,SDAG-X64-NOSSE
; RUN: llc < %s -mtriple=x86_64-- -fast-isel -fast-isel-abort=1 -mattr=-sse | FileCheck %s --check-prefixes=X64-NOSSE,FASTISEL-X64-NOSSE
-; RUN: llc < %s -mtriple=x86_64-- -global-isel -global-isel-abort=2 -mattr=-sse | FileCheck %s --check-prefixes=X64-NOSSE,GISEL-X64-NOSSE
+; RUN: llc < %s -mtriple=x86_64-- -global-isel -global-isel-abort=1 -mattr=-sse | FileCheck %s --check-prefixes=GISEL-X64-NOSSE
; RUN: llc < %s -mtriple=i686-- | FileCheck %s --check-prefixes=X86,SDAG-X86
; RUN: llc < %s -mtriple=i686-- -fast-isel -fast-isel-abort=1 | FileCheck %s --check-prefixes=X86,FASTISEL-X86
-; RUN: llc < %s -mtriple=i686-- -global-isel -global-isel-abort=2 | FileCheck %s --check-prefixes=X86,GISEL-X86
+; RUN: llc < %s -mtriple=i686-- -global-isel -global-isel-abort=1 | FileCheck %s --check-prefixes=GISEL-X86
; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s --check-prefixes=X64,SDAG-X64
; RUN: llc < %s -mtriple=x86_64-- -fast-isel -fast-isel-abort=1 | FileCheck %s --check-prefixes=X64,FASTISEL-X64
-; RUN: llc < %s -mtriple=x86_64-- -global-isel -global-isel-abort=2 | FileCheck %s --check-prefixes=X64,GISEL-X64
+; RUN: llc < %s -mtriple=x86_64-- -global-isel -global-isel-abort=1 | FileCheck %s --check-prefixes=GISEL-X64
declare void @llvm.set.rounding(i32 %x)
@@ -24,6 +24,21 @@ define void @func_01() nounwind {
; X86-NOSSE-NEXT: popl %eax
; X86-NOSSE-NEXT: retl
;
+; GISEL-X86-NOSSE-LABEL: func_01:
+; GISEL-X86-NOSSE: # %bb.0:
+; GISEL-X86-NOSSE-NEXT: pushl %eax
+; GISEL-X86-NOSSE-NEXT: fnstcw (%esp)
+; GISEL-X86-NOSSE-NEXT: movw $-3073, %ax # imm = 0xF3FF
+; GISEL-X86-NOSSE-NEXT: andw (%esp), %ax
+; GISEL-X86-NOSSE-NEXT: movw $201, %cx
+; GISEL-X86-NOSSE-NEXT: shlw $4, %cx
+; GISEL-X86-NOSSE-NEXT: andw $3072, %cx # imm = 0xC00
+; GISEL-X86-NOSSE-NEXT: orw %ax, %cx
+; GISEL-X86-NOSSE-NEXT: movw %cx, (%esp)
+; GISEL-X86-NOSSE-NEXT: fldcw (%esp)
+; GISEL-X86-NOSSE-NEXT: popl %eax
+; GISEL-X86-NOSSE-NEXT: retl
+;
; X64-NOSSE-LABEL: func_01:
; X64-NOSSE: # %bb.0:
; X64-NOSSE-NEXT: fnstcw -{{[0-9]+}}(%rsp)
@@ -31,6 +46,19 @@ define void @func_01() nounwind {
; X64-NOSSE-NEXT: fldcw -{{[0-9]+}}(%rsp)
; X64-NOSSE-NEXT: retq
;
+; GISEL-X64-NOSSE-LABEL: func_01:
+; GISEL-X64-NOSSE: # %bb.0:
+; GISEL-X64-NOSSE-NEXT: fnstcw -{{[0-9]+}}(%rsp)
+; GISEL-X64-NOSSE-NEXT: movw $-3073, %ax # imm = 0xF3FF
+; GISEL-X64-NOSSE-NEXT: andw -{{[0-9]+}}(%rsp), %ax
+; GISEL-X64-NOSSE-NEXT: movw $201, %cx
+; GISEL-X64-NOSSE-NEXT: shlw $4, %cx
+; GISEL-X64-NOSSE-NEXT: andw $3072, %cx # imm = 0xC00
+; GISEL-X64-NOSSE-NEXT: orw %ax, %cx
+; GISEL-X64-NOSSE-NEXT: movw %cx, -{{[0-9]+}}(%rsp)
+; GISEL-X64-NOSSE-NEXT: fldcw -{{[0-9]+}}(%rsp)
+; GISEL-X64-NOSSE-NEXT: retq
+;
; X86-LABEL: func_01:
; X86: # %bb.0:
; X86-NEXT: pushl %eax
@@ -40,6 +68,21 @@ define void @func_01() nounwind {
; X86-NEXT: popl %eax
; X86-NEXT: retl
;
+; GISEL-X86-LABEL: func_01:
+; GISEL-X86: # %bb.0:
+; GISEL-X86-NEXT: pushl %eax
+; GISEL-X86-NEXT: fnstcw (%esp)
+; GISEL-X86-NEXT: movw $-3073, %ax # imm = 0xF3FF
+; GISEL-X86-NEXT: andw (%esp), %ax
+; GISEL-X86-NEXT: movw $201, %cx
+; GISEL-X86-NEXT: shlw $4, %cx
+; GISEL-X86-NEXT: andw $3072, %cx # imm = 0xC00
+; GISEL-X86-NEXT: orw %ax, %cx
+; GISEL-X86-NEXT: movw %cx, (%esp)
+; GISEL-X86-NEXT: fldcw (%esp)
+; GISEL-X86-NEXT: popl %eax
+; GISEL-X86-NEXT: retl
+;
; X64-LABEL: func_01:
; X64: # %bb.0:
; X64-NEXT: fnstcw -{{[0-9]+}}(%rsp)
@@ -49,6 +92,27 @@ define void @func_01() nounwind {
; X64-NEXT: orb $96, -{{[0-9]+}}(%rsp)
; X64-NEXT: ldmxcsr -{{[0-9]+}}(%rsp)
; X64-NEXT: retq
+;
+; GISEL-X64-LABEL: func_01:
+; GISEL-X64: # %bb.0:
+; GISEL-X64-NEXT: fnstcw -{{[0-9]+}}(%rsp)
+; GISEL-X64-NEXT: movw $-3073, %ax # imm = 0xF3FF
+; GISEL-X64-NEXT: andw -{{[0-9]+}}(%rsp), %ax
+; GISEL-X64-NEXT: movw $201, %cx
+; GISEL-X64-NEXT: shlw $4, %cx
+; GISEL-X64-NEXT: andw $3072, %cx # imm = 0xC00
+; GISEL-X64-NEXT: orw %cx, %ax
+; GISEL-X64-NEXT: movw %ax, -{{[0-9]+}}(%rsp)
+; GISEL-X64-NEXT: fldcw -{{[0-9]+}}(%rsp)
+; GISEL-X64-NEXT: stmxcsr -{{[0-9]+}}(%rsp)
+; GISEL-X64-NEXT: movl $-24577, %eax # imm = 0x9FFF
+; GISEL-X64-NEXT: andl -{{[0-9]+}}(%rsp), %eax
+; GISEL-X64-NEXT: movzwl %cx, %ecx
+; GISEL-X64-NEXT: shll $3, %ecx
+; GISEL-X64-NEXT: orl %eax, %ecx
+; GISEL-X64-NEXT: movl %ecx, -{{[0-9]+}}(%rsp)
+; GISEL-X64-NEXT: ldmxcsr -{{[0-9]+}}(%rsp)
+; GISEL-X64-NEXT: retq
call void @llvm.set.rounding(i32 0) ; TowardZero (CW[11-10] = 11)
ret void
}
@@ -63,6 +127,21 @@ define void @func_02() nounwind {
; X86-NOSSE-NEXT: popl %eax
; X86-NOSSE-NEXT: retl
;
+; GISEL-X86-NOSSE-LABEL: func_02:
+; GISEL-X86-NOSSE: # %bb.0:
+; GISEL-X86-NOSSE-NEXT: pushl %eax
+; GISEL-X86-NOSSE-NEXT: fnstcw (%esp)
+; GISEL-X86-NOSSE-NEXT: movw $-3073, %ax # imm = 0xF3FF
+; GISEL-X86-NOSSE-NEXT: andw (%esp), %ax
+; GISEL-X86-NOSSE-NEXT: movw $201, %cx
+; GISEL-X86-NOSSE-NEXT: shlw $6, %cx
+; GISEL-X86-NOSSE-NEXT: andw $3072, %cx # imm = 0xC00
+; GISEL-X86-NOSSE-NEXT: orw %ax, %cx
+; GISEL-X86-NOSSE-NEXT: movw %cx, (%esp)
+; GISEL-X86-NOSSE-NEXT: fldcw (%esp)
+; GISEL-X86-NOSSE-NEXT: popl %eax
+; GISEL-X86-NOSSE-NEXT: retl
+;
; X64-NOSSE-LABEL: func_02:
; X64-NOSSE: # %bb.0:
; X64-NOSSE-NEXT: fnstcw -{{[0-9]+}}(%rsp)
@@ -70,6 +149,19 @@ define void @func_02() nounwind {
; X64-NOSSE-NEXT: fldcw -{{[0-9]+}}(%rsp)
; X64-NOSSE-NEXT: retq
;
+; GISEL-X64-NOSSE-LABEL: func_02:
+; GISEL-X64-NOSSE: # %bb.0:
+; GISEL-X64-NOSSE-NEXT: fnstcw -{{[0-9]+}}(%rsp)
+; GISEL-X64-NOSSE-NEXT: movw $-3073, %ax # imm = 0xF3FF
+; GISEL-X64-NOSSE-NEXT: andw -{{[0-9]+}}(%rsp), %ax
+; GISEL-X64-NOSSE-NEXT: movw $201, %cx
+; GISEL-X64-NOSSE-NEXT: shlw $6, %cx
+; GISEL-X64-NOSSE-NEXT: andw $3072, %cx # imm = 0xC00
+; GISEL-X64-NOSSE-NEXT: orw %ax, %cx
+; GISEL-X64-NOSSE-NEXT: movw %cx, -{{[0-9]+}}(%rsp)
+; GISEL-X64-NOSSE-NEXT: fldcw -{{[0-9]+}}(%rsp)
+; GISEL-X64-NOSSE-NEXT: retq
+;
; X86-LABEL: func_02:
; X86: # %bb.0:
; X86-NEXT: pushl %eax
@@ -79,6 +171,21 @@ define void @func_02() nounwind {
; X86-NEXT: popl %eax
; X86-NEXT: retl
;
+; GISEL-X86-LABEL: func_02:
+; GISEL-X86: # %bb.0:
+; GISEL-X86-NEXT: pushl %eax
+; GISEL-X86-NEXT: fnstcw (%esp)
+; GISEL-X86-NEXT: movw $-3073, %ax # imm = 0xF3FF
+; GISEL-X86-NEXT: andw (%esp), %ax
+; GISEL-X86-NEXT: movw $201, %cx
+; GISEL-X86-NEXT: shlw $6, %cx
+; GISEL-X86-NEXT: andw $3072, %cx # imm = 0xC00
+; GISEL-X86-NEXT: orw %ax, %cx
+; GISEL-X86-NEXT: movw %cx, (%esp)
+; GISEL-X86-NEXT: fldcw (%esp)
+; GISEL-X86-NEXT: popl %eax
+; GISEL-X86-NEXT: retl
+;
; X64-LABEL: func_02:
; X64: # %bb.0:
; X64-NEXT: fnstcw -{{[0-9]+}}(%rsp)
@@ -88,6 +195,27 @@ define void @func_02() nounwind {
; X64-NEXT: andb $-97, -{{[0-9]+}}(%rsp)
; X64-NEXT: ldmxcsr -{{[0-9]+}}(%rsp)
; X64-NEXT: retq
+;
+; GISEL-X64-LABEL: func_02:
+; GISEL-X64: # %bb.0:
+; GISEL-X64-NEXT: fnstcw -{{[0-9]+}}(%rsp)
+; GISEL-X64-NEXT: movw $-3073, %ax # imm = 0xF3FF
+; GISEL-X64-NEXT: andw -{{[0-9]+}}(%rsp), %ax
+; GISEL-X64-NEXT: movw $201, %cx
+; GISEL-X64-NEXT: shlw $6, %cx
+; GISEL-X64-NEXT: andw $3072, %cx # imm = 0xC00
+; GISEL-X64-NEXT: orw %cx, %ax
+; GISEL-X64-NEXT: movw %ax, -{{[0-9]+}}(%rsp)
+; GISEL-X64-NEXT: fldcw -{{[0-9]+}}(%rsp)
+; GISEL-X64-NEXT: stmxcsr -{{[0-9]+}}(%rsp)
+; GISEL-X64-NEXT: movl $-24577, %eax # imm = 0x9FFF
+; GISEL-X64-NEXT: andl -{{[0-9]+}}(%rsp), %eax
+; GISEL-X64-NEXT: movzwl %cx, %ecx
+; GISEL-X64-NEXT: shll $3, %ecx
+; GISEL-X64-NEXT: orl %eax, %ecx
+; GISEL-X64-NEXT: mov...
[truncated]
|
arsenm
approved these changes
Sep 4, 2025
arsenm
approved these changes
Sep 5, 2025
e-kud
reviewed
Sep 7, 2025
e-kud
reviewed
Sep 16, 2025
Contributor
e-kud
left a comment
e-kud
left a comment
There was a problem hiding this comment.
Looks good. SDAG is able to deduce required bit and load only required byte when we load the whole word. Not sure if we need to implement it as a part of lowering.
arsenm
reviewed
Sep 16, 2025
RKSimon
requested changes
Sep 19, 2025
Collaborator
RKSimon
left a comment
RKSimon
left a comment
There was a problem hiding this comment.
please fix the merge conflict
arsenm
reviewed
Sep 25, 2025
arsenm
approved these changes
Sep 25, 2025
Co-authored-by: Matt Arsenault <[email protected]>
Co-authored-by: Matt Arsenault <[email protected]>
Co-authored-by: Matt Arsenault <[email protected]>
Contributor
Author
Contributor
|
The |
Contributor
Author
|
Hi @pranavk, |
Contributor
This should not SIGABRT |
arsenm
pushed a commit
that referenced
this pull request
Sep 26, 2025
#160823) - Fix #156591 (comment) - As per https://cdrdv2.intel.com/v1/dl/getContent/671200 default rounding mode is **round to nearest**.
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Sep 26, 2025
…set.rounding (#160823) - Fix llvm/llvm-project#156591 (comment) - As per https://cdrdv2.intel.com/v1/dl/getContent/671200 default rounding mode is **round to nearest**.
Contributor
|
Yeah, I meant the process crashes in general while it shouldn't. We noticed this because our test runner uses |
mahesh-attarde
pushed a commit
to mahesh-attarde/llvm-project
that referenced
this pull request
Oct 3, 2025
- This implementation is adapted from **SDAG X86TargetLowering::LowerSET_ROUNDING**.
mahesh-attarde
pushed a commit
to mahesh-attarde/llvm-project
that referenced
this pull request
Oct 3, 2025
llvm#160823) - Fix llvm#156591 (comment) - As per https://cdrdv2.intel.com/v1/dl/getContent/671200 default rounding mode is **round to nearest**.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.