From 54ebf1c4a19c7e6556bf193c42250d65120847de Mon Sep 17 00:00:00 2001
From: Phoebe Wang <phoebe.wang@intel.com>
Date: Thu, 1 Dec 2022 22:31:51 +0800
Subject: [PATCH] [X86][FP16] Do not combine fminnum/fmaxnum for FP16 emulation

Under the emulation situation, we lack native fmin/fmax instruction support.

Fixes #59258

Reviewed By: skan, spatel

Differential Revision: https://reviews.llvm.org/D139078
---
 llvm/lib/Target/X86/X86ISelLowering.cpp |   4 +-
 llvm/test/CodeGen/X86/pr59258.ll        | 169 ++++++++++++++++++++++++++++++++
 2 files changed, 171 insertions(+), 2 deletions(-)
 create mode 100644 llvm/test/CodeGen/X86/pr59258.ll

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 9b43df8..3acbe7d 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -51968,12 +51968,12 @@ static SDValue combineFMinFMax(SDNode *N, SelectionDAG &DAG) {
 
 static SDValue combineFMinNumFMaxNum(SDNode *N, SelectionDAG &DAG,
                                      const X86Subtarget &Subtarget) {
-  if (Subtarget.useSoftFloat())
+  EVT VT = N->getValueType(0);
+  if (Subtarget.useSoftFloat() || isSoftFP16(VT, Subtarget))
     return SDValue();
 
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
 
-  EVT VT = N->getValueType(0);
   if (!((Subtarget.hasSSE1() && VT == MVT::f32) ||
         (Subtarget.hasSSE2() && VT == MVT::f64) ||
         (Subtarget.hasFP16() && VT == MVT::f16) ||
diff --git a/llvm/test/CodeGen/X86/pr59258.ll b/llvm/test/CodeGen/X86/pr59258.ll
new file mode 100644
index 0000000..fb2d219
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pr59258.ll
@@ -0,0 +1,169 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s
+
+define <8 x half> @cvt_and_clamp2(<8 x float>) nounwind {
+; CHECK-LABEL: cvt_and_clamp2:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    subq $120, %rsp
+; CHECK-NEXT:    movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movaps %xmm1, %xmm0
+; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3],xmm1[3,3]
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; CHECK-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; CHECK-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    xorps %xmm1, %xmm1
+; CHECK-NEXT:    callq fmaxf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    xorps %xmm1, %xmm1
+; CHECK-NEXT:    callq fmaxf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    xorps %xmm1, %xmm1
+; CHECK-NEXT:    callq fmaxf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    xorps %xmm1, %xmm1
+; CHECK-NEXT:    callq fmaxf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    xorps %xmm1, %xmm1
+; CHECK-NEXT:    callq fmaxf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    xorps %xmm1, %xmm1
+; CHECK-NEXT:    callq fmaxf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    xorps %xmm1, %xmm1
+; CHECK-NEXT:    callq fmaxf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
+; CHECK-NEXT:    movss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    xorps %xmm1, %xmm1
+; CHECK-NEXT:    callq fmaxf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq fminf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq fminf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
+; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq fminf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq fminf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; CHECK-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; CHECK-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
+; CHECK-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq fminf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq fminf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
+; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq fminf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload
+; CHECK-NEXT:    # xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq __extendhfsf2@PLT
+; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    callq fminf@PLT
+; CHECK-NEXT:    callq __truncsfhf2@PLT
+; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; CHECK-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; CHECK-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
+; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; CHECK-NEXT:    # xmm1 = xmm1[0],mem[0]
+; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    addq $120, %rsp
+; CHECK-NEXT:    retq
+    %2 = fptrunc <8 x float> %0 to <8 x half>
+    %3 = call <8 x half> @llvm.maxnum.v8f16(<8 x half> zeroinitializer, <8 x half> %2)
+    %4 = call <8 x half> @llvm.minnum.v8f16(<8 x half> %3, <8 x half> <half 1.0, half 1.0, half 1.0, half 1.0, half 1.0, half 1.0, half 1.0, half 1.0>)
+    ret <8 x half> %4
+}
+
+declare <8 x half> @llvm.maxnum.v8f16(<8 x half>, <8 x half>)
+declare <8 x half> @llvm.minnum.v8f16(<8 x half>, <8 x half>)
-- 
2.7.4