[RISC-V] llvm.*maximum/llvm.*minimum can't be lowered to soft floats

[dcaballe]

Hi there,

It looks like llvm.maximum and llvm.vector.reduce.fmaximum (and I assume the minimum counterparts as well) can't be lowered to soft floats.

Repro:

define float @test(<4 x float> %in, float %in2, float %in3) {
  %mr = tail call float @llvm.vector.reduce.fmaximum.v4f32(<4 x float> %in)
  %m = tail call float @llvm.maximum.f32(float %in2, float %in3)
  %res = fadd float %m, %mr
  ret float %res
}

declare float @llvm.vector.reduce.fmaximum.v4f32(<4 x float>)

declare float @llvm.maximum.f32(float, float)

llc -mattr=+m -mtriple=riscv32-pc-linux-elf -target-abi=ilp32 test.ll

SoftenFloatResult #0: t46: f32 = fmaximum t35, t34

LLVM ERROR: Do not know how to soften the result of this operator!
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: llc -mattr=+m -mtriple=riscv32-pc-linux-elf -target-abi=ilp32 test.ll
1.      Running pass 'Function Pass Manager' on module 'test.ll'.
2.      Running pass 'RISC-V DAG->DAG Pattern Instruction Selection' on function '@test'

[llvmbot]

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

Author: Diego Caballero (dcaballe)

Hi there,

It looks like llvm.maximum and llvm.vector.reduce.fmaximum (and I assume the minimum counterparts as well) can't be lowered to soft floats.

Repro:

define float @<!-- -->test(&lt;4 x float&gt; %in, float %in2, float %in3) {
  %mr = tail call float @<!-- -->llvm.vector.reduce.fmaximum.v4f32(&lt;4 x float&gt; %in)
  %m = tail call float @<!-- -->llvm.maximum.f32(float %in2, float %in3)
  %res = fadd float %m, %mr
  ret float %res
}

declare float @<!-- -->llvm.vector.reduce.fmaximum.v4f32(&lt;4 x float&gt;)

declare float @<!-- -->llvm.maximum.f32(float, float)

llc -mattr=+m -mtriple=riscv32-pc-linux-elf -target-abi=ilp32 test.ll

SoftenFloatResult #<!-- -->0: t46: f32 = fmaximum t35, t34

LLVM ERROR: Do not know how to soften the result of this operator!
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: llc -mattr=+m -mtriple=riscv32-pc-linux-elf -target-abi=ilp32 test.ll
1.      Running pass 'Function Pass Manager' on module 'test.ll'.
2.      Running pass 'RISC-V DAG-&gt;DAG Pattern Instruction Selection' on function '@<!-- -->test'

[llvmbot]

@llvm/issue-subscribers-bug

Author: Diego Caballero (dcaballe)

Hi there,

It looks like llvm.maximum and llvm.vector.reduce.fmaximum (and I assume the minimum counterparts as well) can't be lowered to soft floats.

Repro:

define float @<!-- -->test(&lt;4 x float&gt; %in, float %in2, float %in3) {
  %mr = tail call float @<!-- -->llvm.vector.reduce.fmaximum.v4f32(&lt;4 x float&gt; %in)
  %m = tail call float @<!-- -->llvm.maximum.f32(float %in2, float %in3)
  %res = fadd float %m, %mr
  ret float %res
}

declare float @<!-- -->llvm.vector.reduce.fmaximum.v4f32(&lt;4 x float&gt;)

declare float @<!-- -->llvm.maximum.f32(float, float)

llc -mattr=+m -mtriple=riscv32-pc-linux-elf -target-abi=ilp32 test.ll

SoftenFloatResult #<!-- -->0: t46: f32 = fmaximum t35, t34

LLVM ERROR: Do not know how to soften the result of this operator!
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: llc -mattr=+m -mtriple=riscv32-pc-linux-elf -target-abi=ilp32 test.ll
1.      Running pass 'Function Pass Manager' on module 'test.ll'.
2.      Running pass 'RISC-V DAG-&gt;DAG Pattern Instruction Selection' on function '@<!-- -->test'

[dtcxzyw]

Duplicate of #64206.