From 421dcc59212a73b82141caa2c94ea340a7b34deb Mon Sep 17 00:00:00 2001
From: Daniel Sanders <daniel.sanders@imgtec.com>
Date: Tue, 24 Sep 2013 13:02:08 +0000
Subject: [PATCH] [mips][msa] Added partial support for matching fmax_a from
 normal IR (i.e. not intrinsics)

This covers the case where fmax_a can be used to implement ISD::FABS.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@191296 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/Mips/MipsMSAInstrInfo.td       | 14 +++++++++++
 lib/Target/Mips/MipsSEISelLowering.cpp    |  1 +
 test/CodeGen/Mips/msa/arithmetic_float.ll | 30 +++++++++++++++++++++++
 3 files changed, 45 insertions(+)

diff --git a/lib/Target/Mips/MipsMSAInstrInfo.td b/lib/Target/Mips/MipsMSAInstrInfo.td
index d0dbb152112..959c05aa0a9 100644
--- a/lib/Target/Mips/MipsMSAInstrInfo.td
+++ b/lib/Target/Mips/MipsMSAInstrInfo.td
@@ -2946,6 +2946,20 @@ def ST_FW : MSAPat<(store (v4f32 MSA128W:$ws), addrRegImm:$addr),
 def ST_FD : MSAPat<(store (v2f64 MSA128D:$ws), addrRegImm:$addr),
                    (ST_D MSA128D:$ws, addrRegImm:$addr)>;
 
+class MSA_FABS_PSEUDO_DESC_BASE<RegisterClass RCWD, RegisterClass RCWS = RCWD,
+                                InstrItinClass itin = NoItinerary> :
+  MipsPseudo<(outs RCWD:$wd),
+             (ins RCWS:$ws),
+             [(set RCWD:$wd, (fabs RCWS:$ws))]> {
+  InstrItinClass Itinerary = itin;
+}
+def FABS_W : MSA_FABS_PSEUDO_DESC_BASE<MSA128W>,
+             PseudoInstExpansion<(FMAX_A_W MSA128W:$wd, MSA128W:$ws,
+                                           MSA128W:$ws)>;
+def FABS_D : MSA_FABS_PSEUDO_DESC_BASE<MSA128D>,
+             PseudoInstExpansion<(FMAX_A_D MSA128D:$wd, MSA128D:$ws,
+                                           MSA128D:$ws)>;
+
 class MSABitconvertPat<ValueType DstVT, ValueType SrcVT,
                        RegisterClass DstRC, list<Predicate> preds = [HasMSA]> :
    MSAPat<(DstVT (bitconvert SrcVT:$src)),
diff --git a/lib/Target/Mips/MipsSEISelLowering.cpp b/lib/Target/Mips/MipsSEISelLowering.cpp
index ef2217c56ac..30774540eeb 100644
--- a/lib/Target/Mips/MipsSEISelLowering.cpp
+++ b/lib/Target/Mips/MipsSEISelLowering.cpp
@@ -206,6 +206,7 @@ addMSAFloatType(MVT::SimpleValueType Ty, const TargetRegisterClass *RC) {
   setOperationAction(ISD::EXTRACT_VECTOR_ELT, Ty, Legal);
 
   if (Ty != MVT::v8f16) {
+    setOperationAction(ISD::FABS,  Ty, Legal);
     setOperationAction(ISD::FADD,  Ty, Legal);
     setOperationAction(ISD::FDIV,  Ty, Legal);
     setOperationAction(ISD::FLOG2, Ty, Legal);
diff --git a/test/CodeGen/Mips/msa/arithmetic_float.ll b/test/CodeGen/Mips/msa/arithmetic_float.ll
index 3d26cc72a6d..9487e2cba68 100644
--- a/test/CodeGen/Mips/msa/arithmetic_float.ll
+++ b/test/CodeGen/Mips/msa/arithmetic_float.ll
@@ -128,6 +128,34 @@ define void @fdiv_v2f64(<2 x double>* %c, <2 x double>* %a, <2 x double>* %b) no
   ; CHECK: .size fdiv_v2f64
 }
 
+define void @fabs_v4f32(<4 x float>* %c, <4 x float>* %a) nounwind {
+  ; CHECK: fabs_v4f32:
+
+  %1 = load <4 x float>* %a
+  ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
+  %2 = tail call <4 x float> @llvm.fabs.v4f32 (<4 x float> %1)
+  ; CHECK-DAG: fmax_a.w [[R3:\$w[0-9]+]], [[R1]], [[R1]]
+  store <4 x float> %2, <4 x float>* %c
+  ; CHECK-DAG: st.w [[R3]], 0($4)
+
+  ret void
+  ; CHECK: .size fabs_v4f32
+}
+
+define void @fabs_v2f64(<2 x double>* %c, <2 x double>* %a) nounwind {
+  ; CHECK: fabs_v2f64:
+
+  %1 = load <2 x double>* %a
+  ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
+  %2 = tail call <2 x double> @llvm.fabs.v2f64 (<2 x double> %1)
+  ; CHECK-DAG: fmax_a.d [[R3:\$w[0-9]+]], [[R1]], [[R1]]
+  store <2 x double> %2, <2 x double>* %c
+  ; CHECK-DAG: st.d [[R3]], 0($4)
+
+  ret void
+  ; CHECK: .size fabs_v2f64
+}
+
 define void @fsqrt_v4f32(<4 x float>* %c, <4 x float>* %a) nounwind {
   ; CHECK: fsqrt_v4f32:
 
@@ -156,5 +184,7 @@ define void @fsqrt_v2f64(<2 x double>* %c, <2 x double>* %a) nounwind {
   ; CHECK: .size fsqrt_v2f64
 }
 
+declare <4 x float>  @llvm.fabs.v4f32(<4 x float>  %Val)
+declare <2 x double> @llvm.fabs.v2f64(<2 x double> %Val)
 declare <4 x float>  @llvm.sqrt.v4f32(<4 x float>  %Val)
 declare <2 x double> @llvm.sqrt.v2f64(<2 x double> %Val)
-- 
2.34.1