[llvm] 04619db - [RISCV] Add short forward branch support for `mul` instruction (#166300)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Nov 3 23:08:54 PST 2025
Author: quic_hchandel
Date: 2025-11-04T12:38:50+05:30
New Revision: 04619db16ba840f1af2ad66f1d3eda9dc086a059
URL: https://github.com/llvm/llvm-project/commit/04619db16ba840f1af2ad66f1d3eda9dc086a059
DIFF: https://github.com/llvm/llvm-project/commit/04619db16ba840f1af2ad66f1d3eda9dc086a059.diff
LOG: [RISCV] Add short forward branch support for `mul` instruction (#166300)
Added:
llvm/test/CodeGen/RISCV/short-forward-branch-opt-mul.ll
Modified:
llvm/lib/Target/RISCV/RISCVExpandPseudoInsts.cpp
llvm/lib/Target/RISCV/RISCVFeatures.td
llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
llvm/lib/Target/RISCV/RISCVInstrInfoSFB.td
llvm/test/CodeGen/RISCV/features-info.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/RISCV/RISCVExpandPseudoInsts.cpp b/llvm/lib/Target/RISCV/RISCVExpandPseudoInsts.cpp
index 526675a682d86..b0453fc57c053 100644
--- a/llvm/lib/Target/RISCV/RISCVExpandPseudoInsts.cpp
+++ b/llvm/lib/Target/RISCV/RISCVExpandPseudoInsts.cpp
@@ -131,6 +131,7 @@ bool RISCVExpandPseudo::expandMI(MachineBasicBlock &MBB,
case RISCV::PseudoCCMAXU:
case RISCV::PseudoCCMIN:
case RISCV::PseudoCCMINU:
+ case RISCV::PseudoCCMUL:
case RISCV::PseudoCCADDW:
case RISCV::PseudoCCSUBW:
case RISCV::PseudoCCSLL:
@@ -237,6 +238,7 @@ bool RISCVExpandPseudo::expandCCOp(MachineBasicBlock &MBB,
case RISCV::PseudoCCMIN: NewOpc = RISCV::MIN; break;
case RISCV::PseudoCCMAXU: NewOpc = RISCV::MAXU; break;
case RISCV::PseudoCCMINU: NewOpc = RISCV::MINU; break;
+ case RISCV::PseudoCCMUL: NewOpc = RISCV::MUL; break;
case RISCV::PseudoCCADDI: NewOpc = RISCV::ADDI; break;
case RISCV::PseudoCCSLLI: NewOpc = RISCV::SLLI; break;
case RISCV::PseudoCCSRLI: NewOpc = RISCV::SRLI; break;
diff --git a/llvm/lib/Target/RISCV/RISCVFeatures.td b/llvm/lib/Target/RISCV/RISCVFeatures.td
index cfee6ab22d4ff..5b72334f58d45 100644
--- a/llvm/lib/Target/RISCV/RISCVFeatures.td
+++ b/llvm/lib/Target/RISCV/RISCVFeatures.td
@@ -1856,6 +1856,11 @@ def TuneShortForwardBranchIMinMax
"true", "Enable short forward branch optimization for min,max instructions in Zbb",
[TuneShortForwardBranchOpt]>;
+def TuneShortForwardBranchIMul
+ : SubtargetFeature<"short-forward-branch-i-mul", "HasShortForwardBranchIMul",
+ "true", "Enable short forward branch optimization for mul instruction",
+ [TuneShortForwardBranchOpt]>;
+
// Some subtargets require a S2V transfer buffer to move scalars into vectors.
// FIXME: Forming .vx/.vf/.wx/.wf can reduce register pressure.
def TuneNoSinkSplatOperands
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
index c9df787e0012d..b8ab70bd9e386 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
@@ -1703,6 +1703,7 @@ unsigned getPredicatedOpcode(unsigned Opcode) {
case RISCV::MAXU: return RISCV::PseudoCCMAXU;
case RISCV::MIN: return RISCV::PseudoCCMIN;
case RISCV::MINU: return RISCV::PseudoCCMINU;
+ case RISCV::MUL: return RISCV::PseudoCCMUL;
case RISCV::ADDI: return RISCV::PseudoCCADDI;
case RISCV::SLLI: return RISCV::PseudoCCSLLI;
@@ -1754,6 +1755,9 @@ static MachineInstr *canFoldAsPredicatedOp(Register Reg,
MI->getOpcode() == RISCV::MINU || MI->getOpcode() == RISCV::MAXU))
return nullptr;
+ if (!STI.hasShortForwardBranchIMul() && MI->getOpcode() == RISCV::MUL)
+ return nullptr;
+
// Check if MI can be predicated and folded into the CCMOV.
if (getPredicatedOpcode(MI->getOpcode()) == RISCV::INSTRUCTION_LIST_END)
return nullptr;
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoSFB.td b/llvm/lib/Target/RISCV/RISCVInstrInfoSFB.td
index 5a67a5aaba293..494b1c9f98839 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoSFB.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoSFB.td
@@ -110,6 +110,7 @@ def PseudoCCMAX : SFBALU_rr;
def PseudoCCMIN : SFBALU_rr;
def PseudoCCMAXU : SFBALU_rr;
def PseudoCCMINU : SFBALU_rr;
+def PseudoCCMUL : SFBALU_rr;
def PseudoCCADDI : SFBALU_ri;
def PseudoCCANDI : SFBALU_ri;
diff --git a/llvm/test/CodeGen/RISCV/features-info.ll b/llvm/test/CodeGen/RISCV/features-info.ll
index 988d0490afeb6..cf44af608542c 100644
--- a/llvm/test/CodeGen/RISCV/features-info.ll
+++ b/llvm/test/CodeGen/RISCV/features-info.ll
@@ -137,6 +137,7 @@
; CHECK-NEXT: shifted-zextw-fusion - Enable SLLI+SRLI to be fused when computing (shifted) word zero extension.
; CHECK-NEXT: shlcofideleg - 'Shlcofideleg' (Delegating LCOFI Interrupts to VS-mode).
; CHECK-NEXT: short-forward-branch-i-minmax - Enable short forward branch optimization for min,max instructions in Zbb.
+; CHECK-NEXT: short-forward-branch-i-mul - Enable short forward branch optimization for mul instruction.
; CHECK-NEXT: short-forward-branch-opt - Enable short forward branch optimization.
; CHECK-NEXT: shtvala - 'Shtvala' (htval provides all needed values).
; CHECK-NEXT: shvsatpa - 'Shvsatpa' (vsatp supports all modes supported by satp).
diff --git a/llvm/test/CodeGen/RISCV/short-forward-branch-opt-mul.ll b/llvm/test/CodeGen/RISCV/short-forward-branch-opt-mul.ll
new file mode 100644
index 0000000000000..3f780fddafcce
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/short-forward-branch-opt-mul.ll
@@ -0,0 +1,156 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc < %s -mtriple=riscv32 -mattr=+m | FileCheck %s --check-prefixes=RV32I-M
+; RUN: llc < %s -mtriple=riscv64 -mattr=+m | FileCheck %s --check-prefixes=RV64I-M
+; RUN: llc < %s -mtriple=riscv32 -mattr=+m,+short-forward-branch-opt | \
+; RUN: FileCheck %s --check-prefixes=RV32I-SFB-M
+; RUN: llc < %s -mtriple=riscv64 -mattr=+m,+short-forward-branch-opt | \
+; RUN: FileCheck %s --check-prefixes=RV64I-SFB-M
+; RUN: llc < %s -mtriple=riscv32 -mattr=+m,+short-forward-branch-i-mul | \
+; RUN: FileCheck %s --check-prefixes=RV32I-SFBIMul-M
+; RUN: llc < %s -mtriple=riscv64 -mattr=+m,+short-forward-branch-i-mul | \
+; RUN: FileCheck %s --check-prefixes=RV64I-SFBIMul-M
+
+define i32 @select_example_mul_i32(i32 %a, i32 %b, i1 zeroext %x, i32 %y) {
+; RV32I-M-LABEL: select_example_mul_i32:
+; RV32I-M: # %bb.0: # %entry
+; RV32I-M-NEXT: beqz a2, .LBB0_2
+; RV32I-M-NEXT: # %bb.1:
+; RV32I-M-NEXT: mul a1, a0, a3
+; RV32I-M-NEXT: .LBB0_2: # %entry
+; RV32I-M-NEXT: mv a0, a1
+; RV32I-M-NEXT: ret
+;
+; RV64I-M-LABEL: select_example_mul_i32:
+; RV64I-M: # %bb.0: # %entry
+; RV64I-M-NEXT: beqz a2, .LBB0_2
+; RV64I-M-NEXT: # %bb.1:
+; RV64I-M-NEXT: mulw a1, a0, a3
+; RV64I-M-NEXT: .LBB0_2: # %entry
+; RV64I-M-NEXT: mv a0, a1
+; RV64I-M-NEXT: ret
+;
+; RV32I-SFB-M-LABEL: select_example_mul_i32:
+; RV32I-SFB-M: # %bb.0: # %entry
+; RV32I-SFB-M-NEXT: mul a0, a0, a3
+; RV32I-SFB-M-NEXT: bnez a2, .LBB0_2
+; RV32I-SFB-M-NEXT: # %bb.1: # %entry
+; RV32I-SFB-M-NEXT: mv a0, a1
+; RV32I-SFB-M-NEXT: .LBB0_2: # %entry
+; RV32I-SFB-M-NEXT: ret
+;
+; RV64I-SFB-M-LABEL: select_example_mul_i32:
+; RV64I-SFB-M: # %bb.0: # %entry
+; RV64I-SFB-M-NEXT: mulw a0, a0, a3
+; RV64I-SFB-M-NEXT: bnez a2, .LBB0_2
+; RV64I-SFB-M-NEXT: # %bb.1: # %entry
+; RV64I-SFB-M-NEXT: mv a0, a1
+; RV64I-SFB-M-NEXT: .LBB0_2: # %entry
+; RV64I-SFB-M-NEXT: ret
+;
+; RV32I-SFBIMul-M-LABEL: select_example_mul_i32:
+; RV32I-SFBIMul-M: # %bb.0: # %entry
+; RV32I-SFBIMul-M-NEXT: beqz a2, .LBB0_2
+; RV32I-SFBIMul-M-NEXT: # %bb.1: # %entry
+; RV32I-SFBIMul-M-NEXT: mul a1, a0, a3
+; RV32I-SFBIMul-M-NEXT: .LBB0_2: # %entry
+; RV32I-SFBIMul-M-NEXT: mv a0, a1
+; RV32I-SFBIMul-M-NEXT: ret
+;
+; RV64I-SFBIMul-M-LABEL: select_example_mul_i32:
+; RV64I-SFBIMul-M: # %bb.0: # %entry
+; RV64I-SFBIMul-M-NEXT: mulw a0, a0, a3
+; RV64I-SFBIMul-M-NEXT: bnez a2, .LBB0_2
+; RV64I-SFBIMul-M-NEXT: # %bb.1: # %entry
+; RV64I-SFBIMul-M-NEXT: mv a0, a1
+; RV64I-SFBIMul-M-NEXT: .LBB0_2: # %entry
+; RV64I-SFBIMul-M-NEXT: ret
+entry:
+ %res = mul i32 %a, %y
+ %sel = select i1 %x, i32 %res, i32 %b
+ ret i32 %sel
+}
+
+define i64 @select_example_mul_i64(i64 %a, i64 %b, i1 zeroext %x, i64 %y) {
+; RV32I-M-LABEL: select_example_mul_i64:
+; RV32I-M: # %bb.0: # %entry
+; RV32I-M-NEXT: beqz a4, .LBB1_2
+; RV32I-M-NEXT: # %bb.1:
+; RV32I-M-NEXT: mul a2, a0, a6
+; RV32I-M-NEXT: mulhu a3, a0, a5
+; RV32I-M-NEXT: mul a1, a1, a5
+; RV32I-M-NEXT: add a2, a3, a2
+; RV32I-M-NEXT: add a3, a2, a1
+; RV32I-M-NEXT: mul a2, a0, a5
+; RV32I-M-NEXT: .LBB1_2: # %entry
+; RV32I-M-NEXT: mv a0, a2
+; RV32I-M-NEXT: mv a1, a3
+; RV32I-M-NEXT: ret
+;
+; RV64I-M-LABEL: select_example_mul_i64:
+; RV64I-M: # %bb.0: # %entry
+; RV64I-M-NEXT: beqz a2, .LBB1_2
+; RV64I-M-NEXT: # %bb.1:
+; RV64I-M-NEXT: mul a1, a0, a3
+; RV64I-M-NEXT: .LBB1_2: # %entry
+; RV64I-M-NEXT: mv a0, a1
+; RV64I-M-NEXT: ret
+;
+; RV32I-SFB-M-LABEL: select_example_mul_i64:
+; RV32I-SFB-M: # %bb.0: # %entry
+; RV32I-SFB-M-NEXT: mul a6, a0, a6
+; RV32I-SFB-M-NEXT: mulhu a7, a0, a5
+; RV32I-SFB-M-NEXT: mul a1, a1, a5
+; RV32I-SFB-M-NEXT: mul a0, a0, a5
+; RV32I-SFB-M-NEXT: add a6, a7, a6
+; RV32I-SFB-M-NEXT: beqz a4, .LBB1_2
+; RV32I-SFB-M-NEXT: # %bb.1: # %entry
+; RV32I-SFB-M-NEXT: add a3, a6, a1
+; RV32I-SFB-M-NEXT: .LBB1_2: # %entry
+; RV32I-SFB-M-NEXT: bnez a4, .LBB1_4
+; RV32I-SFB-M-NEXT: # %bb.3: # %entry
+; RV32I-SFB-M-NEXT: mv a0, a2
+; RV32I-SFB-M-NEXT: .LBB1_4: # %entry
+; RV32I-SFB-M-NEXT: mv a1, a3
+; RV32I-SFB-M-NEXT: ret
+;
+; RV64I-SFB-M-LABEL: select_example_mul_i64:
+; RV64I-SFB-M: # %bb.0: # %entry
+; RV64I-SFB-M-NEXT: mul a0, a0, a3
+; RV64I-SFB-M-NEXT: bnez a2, .LBB1_2
+; RV64I-SFB-M-NEXT: # %bb.1: # %entry
+; RV64I-SFB-M-NEXT: mv a0, a1
+; RV64I-SFB-M-NEXT: .LBB1_2: # %entry
+; RV64I-SFB-M-NEXT: ret
+;
+; RV32I-SFBIMul-M-LABEL: select_example_mul_i64:
+; RV32I-SFBIMul-M: # %bb.0: # %entry
+; RV32I-SFBIMul-M-NEXT: mul a6, a0, a6
+; RV32I-SFBIMul-M-NEXT: mulhu a7, a0, a5
+; RV32I-SFBIMul-M-NEXT: mul a1, a1, a5
+; RV32I-SFBIMul-M-NEXT: add a6, a7, a6
+; RV32I-SFBIMul-M-NEXT: beqz a4, .LBB1_2
+; RV32I-SFBIMul-M-NEXT: # %bb.1: # %entry
+; RV32I-SFBIMul-M-NEXT: add a3, a6, a1
+; RV32I-SFBIMul-M-NEXT: .LBB1_2: # %entry
+; RV32I-SFBIMul-M-NEXT: beqz a4, .LBB1_4
+; RV32I-SFBIMul-M-NEXT: # %bb.3: # %entry
+; RV32I-SFBIMul-M-NEXT: mul a2, a0, a5
+; RV32I-SFBIMul-M-NEXT: .LBB1_4: # %entry
+; RV32I-SFBIMul-M-NEXT: mv a0, a2
+; RV32I-SFBIMul-M-NEXT: mv a1, a3
+; RV32I-SFBIMul-M-NEXT: ret
+;
+; RV64I-SFBIMul-M-LABEL: select_example_mul_i64:
+; RV64I-SFBIMul-M: # %bb.0: # %entry
+; RV64I-SFBIMul-M-NEXT: beqz a2, .LBB1_2
+; RV64I-SFBIMul-M-NEXT: # %bb.1: # %entry
+; RV64I-SFBIMul-M-NEXT: mul a1, a0, a3
+; RV64I-SFBIMul-M-NEXT: .LBB1_2: # %entry
+; RV64I-SFBIMul-M-NEXT: mv a0, a1
+; RV64I-SFBIMul-M-NEXT: ret
+entry:
+ %res = mul i64 %a, %y
+ %sel = select i1 %x, i64 %res, i64 %b
+ ret i64 %sel
+}
+
More information about the llvm-commits
mailing list