[RISCV] Add fmin/fmax scalar instructions to isAssociativeAndCommutative

Follow-up patch of D140530. We can add FMIN, FMAX to isAssociativeAndCommutative to increase instruction-level parallelism by the existing MachineCombiner pass. Differential Revision: https://reviews.llvm.org/D140602
llvm · Dec 29, 2022 · af5dd27 · af5dd27
1 parent 002005e
commit af5dd27
Show file tree

Hide file tree

Showing 2 changed files with 91 additions and 1 deletion.
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
@@ -1394,6 +1394,12 @@ bool RISCVInstrInfo::isAssociativeAndCommutative(const MachineInstr &Inst,
   case RISCV::MINU:
   case RISCV::MAX:
   case RISCV::MAXU:
+  case RISCV::FMIN_H:
+  case RISCV::FMIN_S:
+  case RISCV::FMIN_D:
+  case RISCV::FMAX_H:
+  case RISCV::FMAX_S:
+  case RISCV::FMAX_D:
     return true;
   }
 

diff --git a/llvm/test/CodeGen/RISCV/machine-combiner.ll b/llvm/test/CodeGen/RISCV/machine-combiner.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=riscv64 -mattr=+d,+zbb -verify-machineinstrs -mcpu=sifive-u74 \
+; RUN: llc -mtriple=riscv64 -mattr=+d,+zbb,+zfh -verify-machineinstrs -mcpu=sifive-u74 \
 ; RUN: -O1 -riscv-enable-machine-combiner=true < %s | \
 ; RUN: FileCheck %s
 
@@ -988,6 +988,84 @@ define i64 @test_reassoc_max_i64(i64 %a0, i64 %a1, i64 %a2, i64 %a3) {
   ret i64 %t2
 }
 
+define half @test_fmin_f16(half %a0, half %a1, half %a2, half %a3) {
+; CHECK-LABEL: test_fmin_f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fmin.h ft0, fa0, fa1
+; CHECK-NEXT:    fmin.h ft1, fa2, fa3
+; CHECK-NEXT:    fmin.h fa0, ft0, ft1
+; CHECK-NEXT:    ret
+  %t0 = call half @llvm.minnum.f16(half %a0, half %a1)
+  %t1 = call half @llvm.minnum.f16(half %t0, half %a2)
+  %t2 = call half @llvm.minnum.f16(half %t1, half %a3)
+  ret half %t2
+}
+
+define float @test_fmin_f32(float %a0, float %a1, float %a2, float %a3) {
+; CHECK-LABEL: test_fmin_f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fmin.s ft0, fa0, fa1
+; CHECK-NEXT:    fmin.s ft1, fa2, fa3
+; CHECK-NEXT:    fmin.s fa0, ft0, ft1
+; CHECK-NEXT:    ret
+  %t0 = call float @llvm.minnum.f32(float %a0, float %a1)
+  %t1 = call float @llvm.minnum.f32(float %t0, float %a2)
+  %t2 = call float @llvm.minnum.f32(float %t1, float %a3)
+  ret float %t2
+}
+
+define double @test_fmin_f64(double %a0, double %a1, double %a2, double %a3) {
+; CHECK-LABEL: test_fmin_f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fmin.d ft0, fa0, fa1
+; CHECK-NEXT:    fmin.d ft1, fa2, fa3
+; CHECK-NEXT:    fmin.d fa0, ft0, ft1
+; CHECK-NEXT:    ret
+  %t0 = call double @llvm.minnum.f64(double %a0, double %a1)
+  %t1 = call double @llvm.minnum.f64(double %t0, double %a2)
+  %t2 = call double @llvm.minnum.f64(double %t1, double %a3)
+  ret double %t2
+}
+
+define half @test_fmax_f16(half %a0, half %a1, half %a2, half %a3) {
+; CHECK-LABEL: test_fmax_f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fmax.h ft0, fa0, fa1
+; CHECK-NEXT:    fmax.h ft1, fa2, fa3
+; CHECK-NEXT:    fmax.h fa0, ft0, ft1
+; CHECK-NEXT:    ret
+  %t0 = call half @llvm.maxnum.f16(half %a0, half %a1)
+  %t1 = call half @llvm.maxnum.f16(half %t0, half %a2)
+  %t2 = call half @llvm.maxnum.f16(half %t1, half %a3)
+  ret half %t2
+}
+
+define float @test_fmax_f32(float %a0, float %a1, float %a2, float %a3) {
+; CHECK-LABEL: test_fmax_f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fmax.s ft0, fa0, fa1
+; CHECK-NEXT:    fmax.s ft1, fa2, fa3
+; CHECK-NEXT:    fmax.s fa0, ft0, ft1
+; CHECK-NEXT:    ret
+  %t0 = call float @llvm.maxnum.f32(float %a0, float %a1)
+  %t1 = call float @llvm.maxnum.f32(float %t0, float %a2)
+  %t2 = call float @llvm.maxnum.f32(float %t1, float %a3)
+  ret float %t2
+}
+
+define double @test_fmax_f64(double %a0, double %a1, double %a2, double %a3) {
+; CHECK-LABEL: test_fmax_f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    fmax.d ft0, fa0, fa1
+; CHECK-NEXT:    fmax.d ft1, fa2, fa3
+; CHECK-NEXT:    fmax.d fa0, ft0, ft1
+; CHECK-NEXT:    ret
+  %t0 = call double @llvm.maxnum.f64(double %a0, double %a1)
+  %t1 = call double @llvm.maxnum.f64(double %t0, double %a2)
+  %t2 = call double @llvm.maxnum.f64(double %t1, double %a3)
+  ret double %t2
+}
+
 declare i8 @llvm.umin.i8(i8 %a, i8 %b)
 declare i16 @llvm.umin.i16(i16 %a, i16 %b)
 declare i32 @llvm.umin.i32(i32 %a, i32 %b)
@@ -1004,3 +1082,9 @@ declare i8 @llvm.smax.i8(i8 %a, i8 %b)
 declare i16 @llvm.smax.i16(i16 %a, i16 %b)
 declare i32 @llvm.smax.i32(i32 %a, i32 %b)
 declare i64 @llvm.smax.i64(i64 %a, i64 %b)
+declare half @llvm.minnum.f16(half, half)
+declare float @llvm.minnum.f32(float, float)
+declare double @llvm.minnum.f64(double, double)
+declare half @llvm.maxnum.f16(half, half)
+declare float @llvm.maxnum.f32(float, float)
+declare double @llvm.maxnum.f64(double, double)