[llvm] [DAG] - computeKnownFPClass - ISD::FMAXIMUM/ISD::FMINIMUM + ISD::FMAXIMUMNUM/ISD::FMINIMUMNUM (PR #190465)
Kartik Ohlan via llvm-commits
llvm-commits at lists.llvm.org
Sun Apr 19 12:51:28 PDT 2026
https://github.com/Ko496-glitch updated https://github.com/llvm/llvm-project/pull/190465
>From c6b9efd9d8726518750cffa067d0f3eafeaad321 Mon Sep 17 00:00:00 2001
From: Kartik Ohlan <kartik7ohlan at gmail.com>
Date: Sun, 19 Apr 2026 15:48:04 -0400
Subject: [PATCH] Cleaning up the code to resolve the conflicts
---
.../lib/CodeGen/SelectionDAG/SelectionDAG.cpp | 37 ++++
.../AMDGPU/fminimum-fmaximum-nofpclass.ll | 170 ++++++++++++++++++
2 files changed, 207 insertions(+)
create mode 100644 llvm/test/CodeGen/AMDGPU/fminimum-fmaximum-nofpclass.ll
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 86245e4044925..9b9f7fe2b5555 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -6135,6 +6135,43 @@ KnownFPClass SelectionDAG::computeKnownFPClass(SDValue Op,
Known.KnownFPClasses &= ~AssertedClasses;
break;
}
+ case ISD::FMAXNUM:
+ case ISD::FMINNUM:
+ case ISD::FMAXIMUM:
+ case ISD::FMINIMUM:
+ case ISD::FMAXIMUMNUM:
+ case ISD::FMINIMUMNUM: {
+ KnownFPClass KnownRHS = computeKnownFPClass(Op.getOperand(1), DemandedElts,
+ InterestedClasses, Depth + 1);
+ if (KnownRHS.isUnknown())
+ break;
+ KnownFPClass KnownLHS = computeKnownFPClass(Op.getOperand(0), DemandedElts,
+ InterestedClasses, Depth + 1);
+ KnownFPClass::MinMaxKind Kind;
+ switch (Opcode) {
+ case ISD::FMAXNUM:
+ Kind = KnownFPClass::MinMaxKind::maxnum;
+ break;
+ case ISD::FMINNUM:
+ Kind = KnownFPClass::MinMaxKind::minnum;
+ break;
+ case ISD::FMAXIMUM:
+ Kind = KnownFPClass::MinMaxKind::maximum;
+ break;
+ case ISD::FMINIMUM:
+ Kind = KnownFPClass::MinMaxKind::minimum;
+ break;
+ case ISD::FMAXIMUMNUM:
+ Kind = KnownFPClass::MinMaxKind::maximumnum;
+ break;
+ default:
+ Kind = KnownFPClass::MinMaxKind::minimumnum;
+ break;
+ }
+ Known = KnownFPClass::minMaxLike(KnownLHS, KnownRHS, Kind,
+ getDenormalMode(Op.getValueType()));
+ break;
+ }
default:
if (Opcode >= ISD::BUILTIN_OP_END || Opcode == ISD::INTRINSIC_WO_CHAIN ||
Opcode == ISD::INTRINSIC_W_CHAIN || Opcode == ISD::INTRINSIC_VOID) {
diff --git a/llvm/test/CodeGen/AMDGPU/fminimum-fmaximum-nofpclass.ll b/llvm/test/CodeGen/AMDGPU/fminimum-fmaximum-nofpclass.ll
new file mode 100644
index 0000000000000..dde7353f9556d
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/fminimum-fmaximum-nofpclass.ll
@@ -0,0 +1,170 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -global-isel=0 < %s | FileCheck %s --check-prefixes=GCN,GFX8
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -global-isel=0 < %s | FileCheck %s --check-prefixes=GCN,GFX9
+
+
+define float @v_test_known_not_snan_maxnum_input_fmed3_r_i_i_f32(float %a, float %b) #0 {
+; GCN-LABEL: v_test_known_not_snan_maxnum_input_fmed3_r_i_i_f32:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_rcp_f32_e32 v0, v0
+; GCN-NEXT: v_add_f32_e32 v1, 1.0, v1
+; GCN-NEXT: v_max_f32_e32 v0, v0, v1
+; GCN-NEXT: v_med3_f32 v0, v0, 2.0, 4.0
+; GCN-NEXT: s_setpc_b64 s[30:31]
+ %a.nnan.add = fdiv nnan float 1.0, %a, !fpmath !0
+ %b.nnan.add = fadd nnan float %b, 1.0
+ %known.not.snan = call float @llvm.maxnum.f32(float %a.nnan.add, float %b.nnan.add)
+ %max = call float @llvm.maxnum.f32(float %known.not.snan, float 2.0)
+ %med = call float @llvm.minnum.f32(float %max, float 4.0)
+ ret float %med
+}
+
+define float @v_maxnum_possible_nan_lhs_input_fmed3_r_i_i_f32(float %a, float %b) #0 {
+; GFX8-LABEL: v_maxnum_possible_nan_lhs_input_fmed3_r_i_i_f32:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_add_f32_e32 v1, 1.0, v1
+; GFX8-NEXT: v_mul_f32_e32 v0, 1.0, v0
+; GFX8-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX8-NEXT: v_med3_f32 v0, v0, 2.0, 4.0
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_maxnum_possible_nan_lhs_input_fmed3_r_i_i_f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1
+; GFX9-NEXT: v_max_f32_e32 v0, v0, v0
+; GFX9-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX9-NEXT: v_med3_f32 v0, v0, 2.0, 4.0
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+ %b.nnan.add = fadd nnan float %b, 1.0
+ %known.not.snan = call float @llvm.maxnum.f32(float %a, float %b.nnan.add)
+ %max = call float @llvm.maxnum.f32(float %known.not.snan, float 2.0)
+ %med = call float @llvm.minnum.f32(float %max, float 4.0)
+ ret float %med
+}
+
+define float @v_maxnum_possible_nan_rhs_input_fmed3_r_i_i_f32(float %a, float %b) #0 {
+; GFX8-LABEL: v_maxnum_possible_nan_rhs_input_fmed3_r_i_i_f32:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_rcp_f32_e32 v0, v0
+; GFX8-NEXT: v_mul_f32_e32 v1, 1.0, v1
+; GFX8-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX8-NEXT: v_med3_f32 v0, v0, 2.0, 4.0
+; GFX8-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_maxnum_possible_nan_rhs_input_fmed3_r_i_i_f32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_rcp_f32_e32 v0, v0
+; GFX9-NEXT: v_max_f32_e32 v1, v1, v1
+; GFX9-NEXT: v_max_f32_e32 v0, v0, v1
+; GFX9-NEXT: v_med3_f32 v0, v0, 2.0, 4.0
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+ %a.nnan.add = fdiv nnan float 1.0, %a, !fpmath !0
+ %known.not.snan = call float @llvm.maxnum.f32(float %a.nnan.add, float %b)
+ %max = call float @llvm.maxnum.f32(float %known.not.snan, float 2.0)
+ %med = call float @llvm.minnum.f32(float %max, float 4.0)
+ ret float %med
+}
+
+define amdgpu_kernel void @v_nnan_inputs_med3_f32_pat0(ptr addrspace(1) %out, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr, ptr addrspace(1) %cptr) {
+; GFX8-LABEL: v_nnan_inputs_med3_f32_pat0:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_load_dwordx8 s[0:7], s[8:9], 0x0
+; GFX8-NEXT: v_lshlrev_b32_e32 v6, 2, v0
+; GFX8-NEXT: s_add_i32 s12, s12, s17
+; GFX8-NEXT: s_mov_b32 flat_scratch_lo, s13
+; GFX8-NEXT: s_lshr_b32 flat_scratch_hi, s12, 8
+; GFX8-NEXT: s_waitcnt lgkmcnt(0)
+; GFX8-NEXT: v_mov_b32_e32 v1, s3
+; GFX8-NEXT: v_add_u32_e32 v0, vcc, s2, v6
+; GFX8-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT: v_mov_b32_e32 v3, s5
+; GFX8-NEXT: v_add_u32_e32 v2, vcc, s4, v6
+; GFX8-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
+; GFX8-NEXT: v_mov_b32_e32 v5, s7
+; GFX8-NEXT: v_add_u32_e32 v4, vcc, s6, v6
+; GFX8-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
+; GFX8-NEXT: flat_load_dword v7, v[0:1] glc
+; GFX8-NEXT: s_waitcnt vmcnt(0)
+; GFX8-NEXT: flat_load_dword v2, v[2:3] glc
+; GFX8-NEXT: s_waitcnt vmcnt(0)
+; GFX8-NEXT: flat_load_dword v3, v[4:5] glc
+; GFX8-NEXT: s_waitcnt vmcnt(0)
+; GFX8-NEXT: v_mov_b32_e32 v1, s1
+; GFX8-NEXT: v_add_u32_e32 v0, vcc, s0, v6
+; GFX8-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT: v_add_f32_e32 v4, 1.0, v7
+; GFX8-NEXT: v_add_f32_e32 v2, 2.0, v2
+; GFX8-NEXT: v_add_f32_e32 v3, 4.0, v3
+; GFX8-NEXT: v_med3_f32 v2, v4, v2, v3
+; GFX8-NEXT: flat_store_dword v[0:1], v2
+; GFX8-NEXT: s_endpgm
+;
+; GFX9-LABEL: v_nnan_inputs_med3_f32_pat0:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_load_dwordx8 s[0:7], s[8:9], 0x0
+; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-NEXT: global_load_dword v1, v0, s[2:3] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: global_load_dword v2, v0, s[4:5] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: global_load_dword v3, v0, s[6:7] glc
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_add_f32_e32 v1, 1.0, v1
+; GFX9-NEXT: v_add_f32_e32 v2, 2.0, v2
+; GFX9-NEXT: v_add_f32_e32 v3, 4.0, v3
+; GFX9-NEXT: v_med3_f32 v1, v1, v2, v3
+; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT: s_endpgm
+ %tid = call i32 @llvm.amdgcn.workitem.id.x()
+ %gep0 = getelementptr float, ptr addrspace(1) %aptr, i32 %tid
+ %gep1 = getelementptr float, ptr addrspace(1) %bptr, i32 %tid
+ %gep2 = getelementptr float, ptr addrspace(1) %cptr, i32 %tid
+ %outgep = getelementptr float, ptr addrspace(1) %out, i32 %tid
+ %a = load volatile float, ptr addrspace(1) %gep0
+ %b = load volatile float, ptr addrspace(1) %gep1
+ %c = load volatile float, ptr addrspace(1) %gep2
+
+ %a.nnan = fadd nnan float %a, 1.0
+ %b.nnan = fadd nnan float %b, 2.0
+ %c.nnan = fadd nnan float %c, 4.0
+
+ %tmp0 = call float @llvm.minnum.f32(float %a.nnan, float %b.nnan)
+ %tmp1 = call float @llvm.maxnum.f32(float %a.nnan, float %b.nnan)
+ %tmp2 = call float @llvm.minnum.f32(float %tmp1, float %c.nnan)
+ %med3 = call float @llvm.maxnum.f32(float %tmp0, float %tmp2)
+ store float %med3, ptr addrspace(1) %outgep
+ ret void
+}
+
+define float @v_test_nnan_input_fmed3_r_i_i_f32_maximum_minimum(float %a) {
+; GCN-LABEL: v_test_nnan_input_fmed3_r_i_i_f32_maximum_minimum:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_add_f32_e32 v0, 1.0, v0
+; GCN-NEXT: v_med3_f32 v0, v0, 2.0, 4.0
+; GCN-NEXT: s_setpc_b64 s[30:31]
+ %a.add = fadd nnan float %a, 1.0
+ %max = call float @llvm.maximum.f32(float %a.add, float 2.0)
+ %med = call float @llvm.minimum.f32(float %max, float 4.0)
+ ret float %med
+}
+
+declare float @llvm.maxnum.f32(float, float) #1
+declare float @llvm.minnum.f32(float, float) #1
+declare float @llvm.maximum.f32(float, float) #1
+declare float @llvm.minimum.f32(float, float) #1
+declare i32 @llvm.amdgcn.workitem.id.x() #1
+
+attributes #0 = { nounwind denormal_fpenv(float: preservesign) }
+attributes #1 = { nounwind readnone speculatable }
+
+!0 = !{float 2.500000e+00}
+
+
+
More information about the llvm-commits
mailing list