AMDGPU/GlobalISel: Legalize s16->s64 G_FPTOSI/G_FPTOUI

Add narrowScalarFor action.
Add narrow scalar for typeIndex == 0 for G_FPTOSI/G_FPTOUI.
Legalize using narrowScalarFor as s16->s32 G_FPTOSI/G_FPTOUI
followed by s32->s64 G_SEXT/G_ZEXT.

Differential Revision: https://reviews.llvm.org/D84010

diff --git a/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h b/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h
index 61e0418..99c08df 100644 (file)
--- a/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h
@@ -696,6 +696,13 @@ public:
     markAllIdxsAsCovered();
     return actionIf(LegalizeAction::NarrowScalar, Predicate, Mutation);
   }
+  /// Narrow the scalar, specified in mutation, when type indexes 0 and 1 is any
+  /// type pair in the given list.
+  LegalizeRuleSet &
+  narrowScalarFor(std::initializer_list<std::pair<LLT, LLT>> Types,
+                  LegalizeMutation Mutation) {
+    return actionFor(LegalizeAction::NarrowScalar, Types, Mutation);
+  }
 
   /// Add more elements to reach the type selected by the mutation if the
   /// predicate is true.

diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index 5b6937e..08fd0b8 100644 (file)
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -1212,6 +1212,22 @@ LegalizerHelper::LegalizeResult LegalizerHelper::narrowScalar(MachineInstr &MI,
     Observer.changedInstr(MI);
     return Legalized;
   }
+  case TargetOpcode::G_FPTOUI: {
+    if (TypeIdx != 0)
+      return UnableToLegalize;
+    Observer.changingInstr(MI);
+    narrowScalarDst(MI, NarrowTy, 0, TargetOpcode::G_ZEXT);
+    Observer.changedInstr(MI);
+    return Legalized;
+  }
+  case TargetOpcode::G_FPTOSI: {
+    if (TypeIdx != 0)
+      return UnableToLegalize;
+    Observer.changingInstr(MI);
+    narrowScalarDst(MI, NarrowTy, 0, TargetOpcode::G_SEXT);
+    Observer.changedInstr(MI);
+    return Legalized;
+  }
   }
 }
 

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 92ff345..415b707 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -654,7 +654,8 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
 
   auto &FPToI = getActionDefinitionsBuilder({G_FPTOSI, G_FPTOUI})
     .legalFor({{S32, S32}, {S32, S64}, {S32, S16}})
-    .customFor({{S64, S64}});
+    .customFor({{S64, S64}})
+    .narrowScalarFor({{S64, S16}}, changeTo(0, S32));
   if (ST.has16BitInsts())
     FPToI.legalFor({{S16, S16}});
   else

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir
index 81056ce..c88ef0d 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir
@@ -587,3 +587,64 @@ body: |
     %1:_(<2 x s64>) = G_FPTOSI %0
     $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
 ...
+
+---
+name: test_fptosi_s16_to_s64
+body: |
+  bb.0:
+    liveins: $vgpr0
+
+    ; SI-LABEL: name: test_fptosi_s16_to_s64
+    ; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; SI: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+    ; SI: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI]](s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[SEXT]](s64)
+    ; VI-LABEL: name: test_fptosi_s16_to_s64
+    ; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; VI: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+    ; VI: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI]](s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[SEXT]](s64)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s16) = G_TRUNC %0
+    %2:_(s64) = G_FPTOSI %1
+    $vgpr0_vgpr1 = COPY %2
+...
+
+---
+name: test_fptosi_v2s16_to_v2s64
+body: |
+  bb.0:
+    liveins: $vgpr0
+
+    ; SI-LABEL: name: test_fptosi_v2s16_to_v2s64
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
+    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; SI: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+    ; SI: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI]](s32)
+    ; SI: [[FPTOSI1:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC1]](s16)
+    ; SI: [[SEXT1:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI1]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[SEXT]](s64), [[SEXT1]](s64)
+    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; VI-LABEL: name: test_fptosi_v2s16_to_v2s64
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; VI: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC]](s16)
+    ; VI: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI]](s32)
+    ; VI: [[FPTOSI1:%[0-9]+]]:_(s32) = G_FPTOSI [[TRUNC1]](s16)
+    ; VI: [[SEXT1:%[0-9]+]]:_(s64) = G_SEXT [[FPTOSI1]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[SEXT]](s64), [[SEXT1]](s64)
+    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s64>) = G_FPTOSI %0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
+...

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir
index b895a1a..97bf448 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir
@@ -757,3 +757,64 @@ body: |
     %1:_(<2 x s64>) = G_FPTOUI %0
     $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
 ...
+
+---
+name: test_fptoui_s16_to_s64
+body: |
+  bb.0:
+    liveins: $vgpr0
+
+    ; SI-LABEL: name: test_fptoui_s16_to_s64
+    ; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; SI: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+    ; SI: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI]](s32)
+    ; SI: $vgpr0_vgpr1 = COPY [[ZEXT]](s64)
+    ; VI-LABEL: name: test_fptoui_s16_to_s64
+    ; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; VI: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+    ; VI: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI]](s32)
+    ; VI: $vgpr0_vgpr1 = COPY [[ZEXT]](s64)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s16) = G_TRUNC %0
+    %2:_(s64) = G_FPTOUI %1
+    $vgpr0_vgpr1 = COPY %2
+...
+
+---
+name: test_fptoui_v2s16_to_v2s64
+body: |
+  bb.0:
+    liveins: $vgpr0
+
+    ; SI-LABEL: name: test_fptoui_v2s16_to_v2s64
+    ; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
+    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; SI: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+    ; SI: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI]](s32)
+    ; SI: [[FPTOUI1:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC1]](s16)
+    ; SI: [[ZEXT1:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI1]](s32)
+    ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[ZEXT]](s64), [[ZEXT1]](s64)
+    ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; VI-LABEL: name: test_fptoui_v2s16_to_v2s64
+    ; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; VI: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC]](s16)
+    ; VI: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI]](s32)
+    ; VI: [[FPTOUI1:%[0-9]+]]:_(s32) = G_FPTOUI [[TRUNC1]](s16)
+    ; VI: [[ZEXT1:%[0-9]+]]:_(s64) = G_ZEXT [[FPTOUI1]](s32)
+    ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[ZEXT]](s64), [[ZEXT1]](s64)
+    ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s64>) = G_FPTOUI %0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
+...