[GlobalISel][AArch64] Select G_FFLOOR

[paquette]

Summary

This teaches the legalizer about G_FFLOOR, and lets us select G_FFLOOR in AArch64.

It updates the existing floating point tests, and adds a select-floor.mir test.

[aemerson]

LGTM.

lib/CodeGen/GlobalISel/LegalizerHelper.cpp

LegalizerHelper::widenScalar(MachineInstr &MI, unsigned TypeIdx, LLT WideTy) {
   case TargetOpcode::G_FMUL:
   case TargetOpcode::G_FSUB:
   case TargetOpcode::G_FMA:
   case TargetOpcode::G_FNEG:
   case TargetOpcode::G_FABS:
   case TargetOpcode::G_FDIV:
   case TargetOpcode::G_FREM:
   case TargetOpcode::G_FCEIL:
+  case TargetOpcode::G_FFLOOR:
   case TargetOpcode::G_FCOS:
   case TargetOpcode::G_FSIN:
   case TargetOpcode::G_FLOG10:
   case TargetOpcode::G_FLOG:
   case TargetOpcode::G_FLOG2:
   case TargetOpcode::G_FSQRT:
   case TargetOpcode::G_FEXP:
     assert(TypeIdx == 0);
LegalizerHelper::fewerElementsVector(MachineInstr &MI, unsigned TypeIdx,
   case G_FMA:
   case G_FPOW:
   case G_FEXP:
   case G_FEXP2:
   case G_FLOG:
   case G_FLOG2:
   case G_FLOG10:
   case G_FCEIL:
+  case G_FFLOOR:
   case G_INTRINSIC_ROUND:
   case G_INTRINSIC_TRUNC:
   case G_FCOS:
   case G_FSIN:
   case G_FSQRT:
     return fewerElementsVectorBasic(MI, TypeIdx, NarrowTy);
   case G_ZEXT:
   case G_SEXT:

lib/Target/AArch64/AArch64LegalizerInfo.cpp

AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST) {
   getActionDefinitionsBuilder({G_UADDE, G_USUBE, G_SADDO, G_SSUBO})
       .legalFor({{s32, s1}, {s64, s1}});
 
   getActionDefinitionsBuilder({G_FADD, G_FSUB, G_FMA, G_FMUL, G_FDIV, G_FNEG})
     .legalFor({s32, s64, v2s64, v4s32, v2s32});
 
   getActionDefinitionsBuilder({G_FREM, G_FPOW}).libcallFor({s32, s64});
 
-  getActionDefinitionsBuilder({G_FCEIL, G_FABS, G_FSQRT})
+  getActionDefinitionsBuilder({G_FCEIL, G_FABS, G_FSQRT, G_FFLOOR})
       // If we don't have full FP16 support, then scalarize the elements of
       // vectors containing fp16 types.
       .fewerElementsIf(
           [=, &ST](const LegalityQuery &Query) {
             const auto &Ty = Query.Types[0];
             return Ty.isVector() && Ty.getElementType() == s16 &&
                    !ST.hasFullFP16();
           },

lib/Target/AArch64/AArch64RegisterBankInfo.cpp

static bool isPreISelGenericFloatingPointOpcode(unsigned Opc) {
   case TargetOpcode::G_FADD:
   case TargetOpcode::G_FSUB:
   case TargetOpcode::G_FMUL:
   case TargetOpcode::G_FDIV:
   case TargetOpcode::G_FCONSTANT:
   case TargetOpcode::G_FPEXT:
   case TargetOpcode::G_FPTRUNC:
   case TargetOpcode::G_FCEIL:
+  case TargetOpcode::G_FFLOOR:
   case TargetOpcode::G_FNEG:
   case TargetOpcode::G_FCOS:
   case TargetOpcode::G_FSIN:
   case TargetOpcode::G_FLOG10:
   case TargetOpcode::G_FLOG:
   case TargetOpcode::G_FLOG2:
   case TargetOpcode::G_FSQRT:
   case TargetOpcode::G_FABS:

test/CodeGen/AArch64/GlobalISel/select-floor.mir

+# RUN: llc -verify-machineinstrs -mtriple aarch64--- \
+# RUN: -run-pass=instruction-select -mattr=+fullfp16 -global-isel %s -o - \
+# RUN: | FileCheck %s
+...
+---
+name:            floor_float
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: fpr }
+  - { id: 1, class: fpr }
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name:            floor_float
+    ; CHECK: %{{[0-9]+}}:fpr32 = FRINTMSr %{{[0-9]+}}
+    liveins: $s0
+    %0:fpr(s32) = COPY $s0
+    %1:fpr(s32) = G_FFLOOR %0
+    $s0 = COPY %1(s32)
+
+...
+---
+name:            floor_double
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: fpr }
+  - { id: 1, class: fpr }
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name:            floor_double
+    ; CHECK: %{{[0-9]+}}:fpr64 = FRINTMDr %{{[0-9]+}}
+    liveins: $d0
+    %0:fpr(s64) = COPY $d0
+    %1:fpr(s64) = G_FFLOOR %0
+    $d0 = COPY %1(s64)
+
+...
+---
+name:            floor_v2f32
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: fpr }
+  - { id: 1, class: fpr }
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name:            floor_v2f32
+    ; CHECK: %{{[0-9]+}}:fpr64 = FRINTMv2f32 %{{[0-9]+}}
+    liveins: $d0
+    %0:fpr(<2 x s32>) = COPY $d0
+    %1:fpr(<2 x s32>) = G_FFLOOR %0
+    $d0 = COPY %1(<2 x s32>)
+
+...
+---
+name:            floor_v4f32
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: fpr }
+  - { id: 1, class: fpr }
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name:            floor_v4f32
+    ; CHECK: %{{[0-9]+}}:fpr128 = FRINTMv4f32 %{{[0-9]+}}
+    liveins: $q0
+    %0:fpr(<4 x s32>) = COPY $q0
+    %1:fpr(<4 x s32>) = G_FFLOOR %0
+    $q0 = COPY %1(<4 x s32>)
+
+...
+---
+name:            floor_v2f64
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: fpr }
+  - { id: 1, class: fpr }
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name:            floor_v2f64
+    ; CHECK: %{{[0-9]+}}:fpr128 = FRINTMv2f64 %{{[0-9]+}}
+    liveins: $q0
+    %0:fpr(<2 x s64>) = COPY $q0
+    %1:fpr(<2 x s64>) = G_FFLOOR %0
+    $q0 = COPY %1(<2 x s64>)
+
+...
+---
+name:            floor_v4f16
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: fpr }
+  - { id: 1, class: fpr }
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name:            floor_v4f16
+    ; CHECK: %{{[0-9]+}}:fpr64 = FRINTMv4f16 %{{[0-9]+}}
+    liveins: $d0
+    %0:fpr(<4 x s16>) = COPY $d0
+    %1:fpr(<4 x s16>) = G_FFLOOR %0
+    $d0 = COPY %1(<4 x s16>)
+
+...
+---
+name:            floor_v8f16
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: fpr }
+  - { id: 1, class: fpr }
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name:            floor_v8f16
+    ; CHECK: %{{[0-9]+}}:fpr128 = FRINTMv8f16 %{{[0-9]+}}
+    liveins: $q0
+    %0:fpr(<8 x s16>) = COPY $q0
+    %1:fpr(<8 x s16>) = G_FFLOOR %0
+    $q0 = COPY %1(<8 x s16>)
+
+...

test/CodeGen/AArch64/arm64-vfloatintrinsics.ll

define %v4f16 @test_v4f16.fabs(%v4f16 %a) {
   ; GISEL-LABEL:          test_v4f16.fabs:
   ; GISEL-NOFP16-COUNT-4: fabs s{{[0-9]+}}, s{{[0-9]+}}
   ; GISEL-FP16-NOT:       fcvt
   ; GISEL-FP16:           fabs.4h
   ; GISEL-FP16-NEXT:      ret
   %1 = call %v4f16 @llvm.fabs.v4f16(%v4f16 %a)
   ret %v4f16 %1
 }
+
+; FALLBACK-NOT: remark{{.*}}test_v4f16.floor
 define %v4f16 @test_v4f16.floor(%v4f16 %a) {
   ; CHECK-LABEL:          test_v4f16.floor:
   ; CHECK-NOFP16-COUNT-4: frintm s{{[0-9]+}}, s{{[0-9]+}}
   ; CHECK-FP16-NOT:       fcvt
   ; CHECK-FP16:           frintm.4h
   ; CHECK-FP16-NEXT:      ret
+
+  ; GISEL-LABEL:          test_v4f16.floor:
+  ; GISEL-NOFP16-COUNT-4: frintm s{{[0-9]+}}, s{{[0-9]+}}
+  ; GISEL-FP16-NOT:       fcvt
+  ; GISEL-FP16:           frintm.4h
+  ; GISEL-FP16-NEXT:      ret
   %1 = call %v4f16 @llvm.floor.v4f16(%v4f16 %a)
   ret %v4f16 %1
 }
 define %v4f16 @test_v4f16.ceil(%v4f16 %a) {
   ; CHECK-LABEL:          test_v4f16.ceil:
   ; CHECK-NOFP16-COUNT-4: frintp s{{[0-9]+}}, s{{[0-9]+}}
   ; CHECK-FP16-NOT:       fcvt
   ; CHECK-FP16:           frintp.4h
define %v8f16 @test_v8f16.fabs(%v8f16 %a) {
   ; GISEL-LABEL:          test_v8f16.fabs:
   ; GISEL-NOFP16-COUNT-8: fabs s{{[0-9]+}}, s{{[0-9]+}}
   ; GISEL-FP16-NOT:       fcvt
   ; GISEL-FP16:           fabs.8h
   ; GISEL-FP16-NEXT:      ret
   %1 = call %v8f16 @llvm.fabs.v8f16(%v8f16 %a)
   ret %v8f16 %1
 }
+
+; FALLBACK-NOT: remark{{.*}}test_v8f16.floor
 define %v8f16 @test_v8f16.floor(%v8f16 %a) {
   ; CHECK-LABEL:          test_v8f16.floor:
   ; CHECK-NOFP16-COUNT-8: frintm s{{[0-9]+}}, s{{[0-9]+}}
   ; CHECK-FP16-NOT:       fcvt
   ; CHECK-FP16:           frintm.8h
   ; CHECK-FP16-NEXT:      ret
+
+  ; GISEL-LABEL:          test_v8f16.floor:
+  ; GISEL-NOFP16-COUNT-8: frintm s{{[0-9]+}}, s{{[0-9]+}}
+  ; GISEL-FP16-NOT:       fcvt
+  ; GISEL-FP16:           frintm.8h
+  ; GISEL-FP16-NEXT:      ret
   %1 = call %v8f16 @llvm.floor.v8f16(%v8f16 %a)
   ret %v8f16 %1
 }
 define %v8f16 @test_v8f16.ceil(%v8f16 %a) {
   ; CHECK-LABEL:          test_v8f16.ceil:
   ; CHECK-NOFP16-COUNT-8: frintp s{{[0-9]+}}, s{{[0-9]+}}
   ; CHECK-FP16-NOT:       fcvt
   ; CHECK-FP16:           frintp.8h

 ; CHECK-LABEL: test_v2f32.fabs:
 ; GISEL-LABEL: test_v2f32.fabs:
 define %v2f32 @test_v2f32.fabs(%v2f32 %a) {
   ; CHECK: fabs.2s
   ; GISEL: fabs.2s
   %1 = call %v2f32 @llvm.fabs.v2f32(%v2f32 %a)
   ret %v2f32 %1
 }
+
+; FALLBACK-NOT: remark{{.*}}test_v2f32.floor
 ; CHECK-LABEL: test_v2f32.floor:
+; GISEL-LABEL: test_v2f32.floor:
 define %v2f32 @test_v2f32.floor(%v2f32 %a) {
   ; CHECK: frintm.2s
+  ; GISEL: frintm.2s
   %1 = call %v2f32 @llvm.floor.v2f32(%v2f32 %a)
   ret %v2f32 %1
 }
 ; CHECK-LABEL: test_v2f32.ceil:
 ; FALLBACK-NOT: remark{{.*}}test_v2f32.ceil
 ; GISEL-LABEL: test_v2f32.ceil:
 define %v2f32 @test_v2f32.ceil(%v2f32 %a) {
   ; CHECK: frintp.2s

 ; CHECK: test_v4f32.fabs:
 ; GISEL: test_v4f32.fabs:
 define %v4f32 @test_v4f32.fabs(%v4f32 %a) {
   ; CHECK: fabs
   ; GISEL: fabs
   %1 = call %v4f32 @llvm.fabs.v4f32(%v4f32 %a)
   ret %v4f32 %1
 }
+
+; FALLBACK-NOT: remark{{.*}}test_v4f32.floor
 ; CHECK: test_v4f32.floor:
+; GISEL: test_v4f32.floor:
 define %v4f32 @test_v4f32.floor(%v4f32 %a) {
   ; CHECK: frintm.4s
+  ; GISEL frintm.4s
   %1 = call %v4f32 @llvm.floor.v4f32(%v4f32 %a)
   ret %v4f32 %1
 }
 ; CHECK: test_v4f32.ceil:
 ; FALLBACK-NOT: remark{{.*}}test_v4f32.ceil
 ; GISEL-LABEL: test_v4f32.ceil:
 define %v4f32 @test_v4f32.ceil(%v4f32 %a) {
   ; CHECK: frintp.4s

 ; CHECK: test_v2f64.fabs:
 ; GISEL: test_v2f64.fabs:
 define %v2f64 @test_v2f64.fabs(%v2f64 %a) {
   ; CHECK: fabs
   ; GISEL: fabs
   %1 = call %v2f64 @llvm.fabs.v2f64(%v2f64 %a)
   ret %v2f64 %1
 }
+
+; FALLBACK-NOT: remark{{.*}}test_v2f64.floor
 ; CHECK: test_v2f64.floor:
+; GISEL: test_v2f64.floor:
 define %v2f64 @test_v2f64.floor(%v2f64 %a) {
   ; CHECK: frintm.2d
+  ; GISEL: frintm.2d
   %1 = call %v2f64 @llvm.floor.v2f64(%v2f64 %a)
   ret %v2f64 %1
 }
 ; CHECK: test_v2f64.ceil:
 ; FALLBACK-NOT: remark{{.*}}test_v2f64.ceil
 ; GISEL-LABEL: test_v2f64.ceil:
 define %v2f64 @test_v2f64.ceil(%v2f64 %a) {
   ; CHECK: frintp.2d

test/CodeGen/AArch64/f16-instructions.ll

 ; CHECK-CVT-NEXT: frintm [[INT32:s[0-9]+]], [[FLOAT32]]
 ; CHECK-CVT-NEXT: fcvt h0, [[INT32]]
 ; CHECK-CVT-NEXT: ret
 
 ; CHECK-FP16-LABEL: test_floor:
 ; CHECK-FP16-NEXT: frintm h0, h0
 ; CHECK-FP16-NEXT: ret
 
+; FALLBACK-NOT: remark:{{.*}}test_floor
+; FALLBACK-FP16-NOT: remark:{{.*}}test_floor
+
+; GISEL-CVT-LABEL: test_floor:
+; GISEL-CVT-NEXT: fcvt [[FLOAT32:s[0-9]+]], h0
+; GISEL-CVT-NEXT: frintm [[INT32:s[0-9]+]], [[FLOAT32]]
+; GISEL-CVT-NEXT: fcvt h0, [[INT32]]
+; GISEL-CVT-NEXT: ret
+
+; GISEL-FP16-LABEL: test_floor:
+; GISEL-FP16-NEXT: frintm h0, h0
+; GISEL-FP16-NEXT: ret
+
 define half @test_floor(half %a) #0 {
   %r = call half @llvm.floor.f16(half %a)
   ret half %r
 }
 
 ; CHECK-CVT-LABEL: test_ceil:
 ; CHECK-CVT-NEXT: fcvt [[FLOAT32:s[0-9]+]], h0
 ; CHECK-CVT-NEXT: frintp [[INT32:s[0-9]+]], [[FLOAT32]]