[MIPS GlobalISel] Select MSA vector generic and builtin fabs

selectImpl is able to select G_FABS when we set bank for vector
operands to fprb. Add detailed tests.
Note: G_FABS is generated from llvm-ir intrinsics llvm.fabs.*,
and at the moment MIPS is not able to generate this intrinsic for
vector type (some targets generate vector llvm.fabs.* from calls
to a builtin function).
We can handle fabs using __builtin_msa_fmax_a_<format> and passing
same vector as both arguments. __builtin_msa_fmax_a_<format> will
be directly selected into FMAX_A_<format> in legalizeIntrinsic.

Differential Revision: https://reviews.llvm.org/D69346

Author: Petar Avramovic

llvm/lib/Target/Mips/MipsLegalizerInfo.cpp

@@ -188,10 +188,10 @@ MipsLegalizerInfo::MipsLegalizerInfo(const MipsSubtarget &ST) {
   getActionDefinitionsBuilder(G_FCONSTANT)
       .legalFor({s32, s64});
 
-  getActionDefinitionsBuilder({G_FABS, G_FSQRT})
+  getActionDefinitionsBuilder(G_FSQRT)
       .legalFor({s32, s64});
 
-  getActionDefinitionsBuilder({G_FADD, G_FSUB, G_FMUL, G_FDIV})
+  getActionDefinitionsBuilder({G_FADD, G_FSUB, G_FMUL, G_FDIV, G_FABS})
       .legalIf([=, &ST](const LegalityQuery &Query) {
         if (CheckTyN(0, Query, {s32, s64}))
           return true;
@@ -425,6 +425,10 @@ bool MipsLegalizerInfo::legalizeIntrinsic(MachineInstr &MI,
   case Intrinsic::mips_fdiv_w:
   case Intrinsic::mips_fdiv_d:
     return MSA3OpIntrinsicToGeneric(MI, TargetOpcode::G_FDIV, MIRBuilder, ST);
+  case Intrinsic::mips_fmax_a_w:
+    return SelectMSA3OpIntrinsic(MI, Mips::FMAX_A_W, MIRBuilder, ST);
+  case Intrinsic::mips_fmax_a_d:
+    return SelectMSA3OpIntrinsic(MI, Mips::FMAX_A_D, MIRBuilder, ST);
   default:
     break;
   }

llvm/lib/Target/Mips/MipsRegisterBankInfo.cpp

@@ -539,14 +539,14 @@ MipsRegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
                                           &Mips::ValueMappings[Mips::GPRIdx]});
     MappingID = CustomMappingID;
     break;
-  case G_FABS:
   case G_FSQRT:
     OperandsMapping = getFprbMapping(Op0Size);
     break;
   case G_FADD:
   case G_FSUB:
   case G_FMUL:
   case G_FDIV:
+  case G_FABS:
     OperandsMapping = getFprbMapping(Op0Size);
     if (Op0Size == 128)
       OperandsMapping = getMSAMapping(MF);

llvm/test/CodeGen/Mips/GlobalISel/instruction-select/fabs_vec.mir

@@ -0,0 +1,60 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=mipsel-linux-gnu -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=P5600
+--- |
+
+  define void @fabs_v4f32(<4 x float>* %a, <4 x float>* %c) { entry: ret void }
+  define void @fabs_v2f64(<2 x double>* %a, <2 x double>* %c) { entry: ret void }
+
+...
+---
+name:            fabs_v4f32
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: fabs_v4f32
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:gpr32 = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:gpr32 = COPY $a1
+    ; P5600: [[LD_W:%[0-9]+]]:msa128w = LD_W [[COPY]], 0 :: (load 16 from %ir.a)
+    ; P5600: [[FABS_W:%[0-9]+]]:msa128w = FABS_W [[LD_W]]
+    ; P5600: ST_W [[FABS_W]], [[COPY1]], 0 :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:gprb(p0) = COPY $a0
+    %1:gprb(p0) = COPY $a1
+    %2:fprb(<4 x s32>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:fprb(<4 x s32>) = G_FABS %2
+    G_STORE %3(<4 x s32>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
+---
+name:            fabs_v2f64
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: fabs_v2f64
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:gpr32 = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:gpr32 = COPY $a1
+    ; P5600: [[LD_D:%[0-9]+]]:msa128d = LD_D [[COPY]], 0 :: (load 16 from %ir.a)
+    ; P5600: [[FABS_D:%[0-9]+]]:msa128d = FABS_D [[LD_D]]
+    ; P5600: ST_D [[FABS_D]], [[COPY1]], 0 :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:gprb(p0) = COPY $a0
+    %1:gprb(p0) = COPY $a1
+    %2:fprb(<2 x s64>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:fprb(<2 x s64>) = G_FABS %2
+    G_STORE %3(<2 x s64>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...

llvm/test/CodeGen/Mips/GlobalISel/legalizer/fabs_vec.mir

@@ -0,0 +1,56 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=mipsel-linux-gnu -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -run-pass=legalizer -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=P5600
+--- |
+
+  define void @fabs_v4f32(<4 x float>* %a, <4 x float>* %c) { entry: ret void }
+  define void @fabs_v2f64(<2 x double>* %a, <2 x double>* %c) { entry: ret void }
+
+...
+---
+name:            fabs_v4f32
+alignment:       4
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: fabs_v4f32
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:_(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:_(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FABS:%[0-9]+]]:_(<4 x s32>) = G_FABS [[LOAD]]
+    ; P5600: G_STORE [[FABS]](<4 x s32>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<4 x s32>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<4 x s32>) = G_FABS %2
+    G_STORE %3(<4 x s32>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
+---
+name:            fabs_v2f64
+alignment:       4
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: fabs_v2f64
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:_(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:_(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FABS:%[0-9]+]]:_(<2 x s64>) = G_FABS [[LOAD]]
+    ; P5600: G_STORE [[FABS]](<2 x s64>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<2 x s64>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<2 x s64>) = G_FABS %2
+    G_STORE %3(<2 x s64>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...

llvm/test/CodeGen/Mips/GlobalISel/legalizer/fabs_vec_builtin.mir

@@ -0,0 +1,59 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=mipsel-linux-gnu -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -run-pass=legalizer -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=P5600
+--- |
+
+  declare <4 x float> @llvm.mips.fmax.a.w(<4 x float>, <4 x float>)
+  define void @fabs_v4f32_builtin(<4 x float>* %a, <4 x float>* %c) { entry: ret void }
+
+  declare <2 x double> @llvm.mips.fmax.a.d(<2 x double>, <2 x double>)
+  define void @fabs_v2f64_builtin(<2 x double>* %a, <2 x double>* %c) { entry: ret void }
+
+...
+---
+name:            fabs_v4f32_builtin
+alignment:       4
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: fabs_v4f32_builtin
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:_(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:_(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:msa128w(<4 x s32>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FMAX_A_W:%[0-9]+]]:msa128w(<4 x s32>) = FMAX_A_W [[LOAD]](<4 x s32>), [[LOAD]](<4 x s32>)
+    ; P5600: G_STORE [[FMAX_A_W]](<4 x s32>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<4 x s32>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.mips.fmax.a.w), %2(<4 x s32>), %2(<4 x s32>)
+    G_STORE %3(<4 x s32>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
+---
+name:            fabs_v2f64_builtin
+alignment:       4
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: fabs_v2f64_builtin
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:_(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:_(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:msa128d(<2 x s64>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FMAX_A_D:%[0-9]+]]:msa128d(<2 x s64>) = FMAX_A_D [[LOAD]](<2 x s64>), [[LOAD]](<2 x s64>)
+    ; P5600: G_STORE [[FMAX_A_D]](<2 x s64>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<2 x s64>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<2 x s64>) = G_INTRINSIC intrinsic(@llvm.mips.fmax.a.d), %2(<2 x s64>), %2(<2 x s64>)
+    G_STORE %3(<2 x s64>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...

llvm/test/CodeGen/Mips/GlobalISel/llvm-ir/fabs_vec.ll

@@ -0,0 +1,34 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -O0 -mtriple=mipsel-linux-gnu -global-isel -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -verify-machineinstrs %s -o -| FileCheck %s -check-prefixes=P5600
+
+declare <4 x float>  @llvm.fabs.v4f32(<4 x float>  %Val)
+define void @fabs_v4f32(<4 x float>* %a, <4 x float>* %c) {
+; P5600-LABEL: fabs_v4f32:
+; P5600:       # %bb.0: # %entry
+; P5600-NEXT:    ld.w $w0, 0($4)
+; P5600-NEXT:    fmax_a.w $w0, $w0, $w0
+; P5600-NEXT:    st.w $w0, 0($5)
+; P5600-NEXT:    jr $ra
+; P5600-NEXT:    nop
+entry:
+  %0 = load <4 x float>, <4 x float>* %a, align 16
+  %fabs = call <4 x float> @llvm.fabs.v4f32 (<4 x float> %0)
+  store <4 x float> %fabs, <4 x float>* %c, align 16
+  ret void
+}
+
+declare <2 x double> @llvm.fabs.v2f64(<2 x double> %Val)
+define void @fabs_v2f64(<2 x double>* %a, <2 x double>* %c) {
+; P5600-LABEL: fabs_v2f64:
+; P5600:       # %bb.0: # %entry
+; P5600-NEXT:    ld.d $w0, 0($4)
+; P5600-NEXT:    fmax_a.d $w0, $w0, $w0
+; P5600-NEXT:    st.d $w0, 0($5)
+; P5600-NEXT:    jr $ra
+; P5600-NEXT:    nop
+entry:
+  %0 = load <2 x double>, <2 x double>* %a, align 16
+  %fabs = call <2 x double> @llvm.fabs.v2f64 (<2 x double> %0)
+  store <2 x double> %fabs, <2 x double>* %c, align 16
+  ret void
+}

llvm/test/CodeGen/Mips/GlobalISel/llvm-ir/fabs_vec_builtin.ll

@@ -0,0 +1,35 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -O0 -mtriple=mipsel-linux-gnu -global-isel -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -verify-machineinstrs %s -o -| FileCheck %s -check-prefixes=P5600
+
+declare <4 x float> @llvm.mips.fmax.a.w(<4 x float>, <4 x float>)
+define void @fabs_v4f32_builtin(<4 x float>* %a, <4 x float>* %c) {
+; P5600-LABEL: fabs_v4f32_builtin:
+; P5600:       # %bb.0: # %entry
+; P5600-NEXT:    ld.w $w0, 0($4)
+; P5600-NEXT:    fmax_a.w $w0, $w0, $w0
+; P5600-NEXT:    st.w $w0, 0($5)
+; P5600-NEXT:    jr $ra
+; P5600-NEXT:    nop
+entry:
+  %0 = load <4 x float>, <4 x float>* %a, align 16
+  %1 = tail call <4 x float> @llvm.mips.fmax.a.w(<4 x float> %0, <4 x float> %0)
+  store <4 x float> %1, <4 x float>* %c, align 16
+  ret void
+}
+
+
+declare <2 x double> @llvm.mips.fmax.a.d(<2 x double>, <2 x double>)
+define void @fabs_v2f64_builtin(<2 x double>* %a, <2 x double>* %c) {
+; P5600-LABEL: fabs_v2f64_builtin:
+; P5600:       # %bb.0: # %entry
+; P5600-NEXT:    ld.d $w0, 0($4)
+; P5600-NEXT:    fmax_a.d $w0, $w0, $w0
+; P5600-NEXT:    st.d $w0, 0($5)
+; P5600-NEXT:    jr $ra
+; P5600-NEXT:    nop
+entry:
+  %0 = load <2 x double>, <2 x double>* %a, align 16
+  %1 = tail call <2 x double> @llvm.mips.fmax.a.d(<2 x double> %0, <2 x double> %0)
+  store <2 x double> %1, <2 x double>* %c, align 16
+  ret void
+}

llvm/test/CodeGen/Mips/GlobalISel/regbankselect/fabs_vec.mir

@@ -0,0 +1,58 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=mipsel-linux-gnu -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -run-pass=regbankselect -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=P5600
+--- |
+
+  define void @fabs_v4f32(<4 x float>* %a, <4 x float>* %c) { entry: ret void }
+  define void @fabs_v2f64(<2 x double>* %a, <2 x double>* %c) { entry: ret void }
+
+...
+---
+name:            fabs_v4f32
+alignment:       4
+legalized:       true
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: fabs_v4f32
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:gprb(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:gprb(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:fprb(<4 x s32>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FABS:%[0-9]+]]:fprb(<4 x s32>) = G_FABS [[LOAD]]
+    ; P5600: G_STORE [[FABS]](<4 x s32>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<4 x s32>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<4 x s32>) = G_FABS %2
+    G_STORE %3(<4 x s32>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
+---
+name:            fabs_v2f64
+alignment:       4
+legalized:       true
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: fabs_v2f64
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:gprb(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:gprb(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:fprb(<2 x s64>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FABS:%[0-9]+]]:fprb(<2 x s64>) = G_FABS [[LOAD]]
+    ; P5600: G_STORE [[FABS]](<2 x s64>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<2 x s64>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<2 x s64>) = G_FABS %2
+    G_STORE %3(<2 x s64>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...