-
Notifications
You must be signed in to change notification settings - Fork 14.5k
[RISCV] Handled the uimm9 offset while FrameIndex folding. #149303
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.
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
base: main
Are you sure you want to change the base?
Conversation
@llvm/pr-subscribers-backend-mips @llvm/pr-subscribers-backend-risc-v Author: UmeshKalappa (ukalappa-mips) ChangesReverted the #148779 changes and
Full diff: https://github.com/llvm/llvm-project/pull/149303.diff 3 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
index 0f948b22759fe..46b297771cfd9 100644
--- a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
@@ -2942,8 +2942,8 @@ bool RISCVDAGToDAGISel::SelectAddrRegImm(SDValue Addr, SDValue &Base,
/// Similar to SelectAddrRegImm, except that the offset is restricted to uimm9.
bool RISCVDAGToDAGISel::SelectAddrRegImm9(SDValue Addr, SDValue &Base,
SDValue &Offset) {
- // FIXME: Support FrameIndex. Need to teach eliminateFrameIndex that only
- // a 9-bit immediate can be folded.
+ if (SelectAddrFrameIndex(Addr, Base, Offset))
+ return true;
SDLoc DL(Addr);
MVT VT = Addr.getSimpleValueType();
@@ -2953,8 +2953,8 @@ bool RISCVDAGToDAGISel::SelectAddrRegImm9(SDValue Addr, SDValue &Base,
if (isUInt<9>(CVal)) {
Base = Addr.getOperand(0);
- // FIXME: Support FrameIndex. Need to teach eliminateFrameIndex that only
- // a 9-bit immediate can be folded.
+ if (auto *FIN = dyn_cast<FrameIndexSDNode>(Base))
+ Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), VT);
Offset = CurDAG->getSignedTargetConstant(CVal, DL, VT);
return true;
}
diff --git a/llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp b/llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp
index 540412366026b..4b9ba0f33276e 100644
--- a/llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp
@@ -573,6 +573,9 @@ bool RISCVRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
int64_t Val = Offset.getFixed();
int64_t Lo12 = SignExtend64<12>(Val);
unsigned Opc = MI.getOpcode();
+ int64_t Imm9Val = SignExtend64<9>(Val);
+ auto &Subtarget = MF.getSubtarget<RISCVSubtarget>();
+
if (Opc == RISCV::ADDI && !isInt<12>(Val)) {
// We chose to emit the canonical immediate sequence rather than folding
// the offset into the using add under the theory that doing so doesn't
@@ -585,6 +588,11 @@ bool RISCVRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
(Lo12 & 0b11111) != 0) {
// Prefetch instructions require the offset to be 32 byte aligned.
MI.getOperand(FIOperandNum + 1).ChangeToImmediate(0);
+ } else if ((Opc == RISCV::PREFETCH_I || Opc == RISCV::PREFETCH_R ||
+ Opc == RISCV::PREFETCH_W) &&
+ Subtarget.hasVendorXMIPSCBOP() && !isUInt<9>(Imm9Val)) {
+ // MIPS Prefetch instructions require the offset to be 9 bits encoded.
+ MI.getOperand(FIOperandNum + 1).ChangeToImmediate(0);
} else if ((Opc == RISCV::PseudoRV32ZdinxLD ||
Opc == RISCV::PseudoRV32ZdinxSD) &&
Lo12 >= 2044) {
diff --git a/llvm/test/CodeGen/RISCV/xmips-cbop.ll b/llvm/test/CodeGen/RISCV/xmips-cbop.ll
index cbbd1de13192c..0cfe3e3f0ab7c 100644
--- a/llvm/test/CodeGen/RISCV/xmips-cbop.ll
+++ b/llvm/test/CodeGen/RISCV/xmips-cbop.ll
@@ -49,3 +49,47 @@ define void @prefetch_inst_read(ptr noundef %ptr) nounwind {
tail call void @llvm.prefetch.p0(ptr nonnull %arrayidx, i32 0, i32 0, i32 0)
ret void
}
+
+define void @prefetch_frameindex_test_neg() nounwind {
+; RV32XMIPSPREFETCH-LABEL: prefetch_frameindex_test_neg:
+; RV32XMIPSPREFETCH: # %bb.0:
+; RV32XMIPSPREFETCH-NEXT: addi sp, sp, -512
+; RV32XMIPSPREFETCH-NEXT: addi a0, sp, -32
+; RV32XMIPSPREFETCH-NEXT: mips.pref 8, 0(a0)
+; RV32XMIPSPREFETCH-NEXT: addi sp, sp, 512
+; RV32XMIPSPREFETCH-NEXT: ret
+;
+; RV64XMIPSPREFETCH-LABEL: prefetch_frameindex_test_neg:
+; RV64XMIPSPREFETCH: # %bb.0:
+; RV64XMIPSPREFETCH-NEXT: addi sp, sp, -512
+; RV64XMIPSPREFETCH-NEXT: addi a0, sp, -32
+; RV64XMIPSPREFETCH-NEXT: mips.pref 8, 0(a0)
+; RV64XMIPSPREFETCH-NEXT: addi sp, sp, 512
+; RV64XMIPSPREFETCH-NEXT: ret
+ %data = alloca [128 x i32], align 4
+ %base = bitcast ptr %data to ptr
+ %ptr = getelementptr [128 x i32], ptr %base, i32 0, i32 -8
+ call void @llvm.prefetch(ptr %ptr, i32 0, i32 0, i32 1)
+ ret void
+}
+
+define void @prefetch_frameindex_test() nounwind {
+; RV32XMIPSPREFETCH-LABEL: prefetch_frameindex_test:
+; RV32XMIPSPREFETCH: # %bb.0:
+; RV32XMIPSPREFETCH-NEXT: addi sp, sp, -512
+; RV32XMIPSPREFETCH-NEXT: mips.pref 8, 32(sp)
+; RV32XMIPSPREFETCH-NEXT: addi sp, sp, 512
+; RV32XMIPSPREFETCH-NEXT: ret
+;
+; RV64XMIPSPREFETCH-LABEL: prefetch_frameindex_test:
+; RV64XMIPSPREFETCH: # %bb.0:
+; RV64XMIPSPREFETCH-NEXT: addi sp, sp, -512
+; RV64XMIPSPREFETCH-NEXT: mips.pref 8, 32(sp)
+; RV64XMIPSPREFETCH-NEXT: addi sp, sp, 512
+; RV64XMIPSPREFETCH-NEXT: ret
+ %data = alloca [128 x i32], align 4
+ %base = bitcast ptr %data to ptr
+ %ptr = getelementptr [128 x i32], ptr %base, i32 0, i32 8
+ call void @llvm.prefetch(ptr %ptr, i32 0, i32 0, i32 1)
+ ret void
+}
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
a271bff
to
742147b
Compare
@topperc ,is that ok to commit ? |
@@ -585,6 +588,11 @@ bool RISCVRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, | |||
(Lo12 & 0b11111) != 0) { | |||
// Prefetch instructions require the offset to be 32 byte aligned. | |||
MI.getOperand(FIOperandNum + 1).ChangeToImmediate(0); | |||
} else if ((Opc == RISCV::PREFETCH_I || Opc == RISCV::PREFETCH_R || | |||
Opc == RISCV::PREFETCH_W) && |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This still isn't looking for RISCV::MIPS_PREF
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@topperc the "ukalappa_riscv_frameindex" branch was not updated ...same updated now
@@ -585,6 +587,10 @@ bool RISCVRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, | |||
(Lo12 & 0b11111) != 0) { | |||
// Prefetch instructions require the offset to be 32 byte aligned. | |||
MI.getOperand(FIOperandNum + 1).ChangeToImmediate(0); | |||
} else if (Opc == RISCV::MIPS_PREFETCH && Subtarget.hasVendorXMIPSCBOP() && |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
You don't need to check Subtarget.hasVendorXMIPSCBOP()
. If you see the opcode you can assume it must be enabled.
@topperc ,we traced out the PrologEpilogInsterter with gdb like (gdb) p MI.dump() (gdb) bt here the offset "Val" is 508 ,which triggers the changes and fold the offset with ADDI. |
Sorry I missed that your first test was changed when you pushed the new version. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
Reverted the #148779 changes and