AMDGPU/GlobalISel: Legalize s16->s64 G_FPEXT

Legalize using narrowScalar as s16->s32 G_FPEXT
followed by s32->s64 G_FPEXT.

Differential Revision: https://reviews.llvm.org/D84030

diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index 3471fbe..3ec04d9 100644 (file)
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -1228,6 +1228,13 @@ LegalizerHelper::LegalizeResult LegalizerHelper::narrowScalar(MachineInstr &MI,
     Observer.changedInstr(MI);
     return Legalized;
   }
+  case TargetOpcode::G_FPEXT:
+    if (TypeIdx != 0)
+      return UnableToLegalize;
+    Observer.changingInstr(MI);
+    narrowScalarDst(MI, NarrowTy, 0, TargetOpcode::G_FPEXT);
+    Observer.changedInstr(MI);
+    return Legalized;
   }
 }
 

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 726e158..99b040e 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -598,7 +598,7 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
 
   getActionDefinitionsBuilder(G_FPEXT)
     .legalFor({{S64, S32}, {S32, S16}})
-    .lowerFor({{S64, S16}}) // FIXME: Implement
+    .narrowScalarFor({{S64, S16}}, changeTo(0, S32))
     .scalarize(0);
 
   getActionDefinitionsBuilder(G_FSUB)

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fpext.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fpext.mir
index e946212..c6ba0b6 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fpext.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fpext.mir
@@ -192,3 +192,45 @@ body: |
     %1:_(<4 x s64>) = G_FPEXT %0
     $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY %1
 ...
+
+---
+name: test_fpext_f16_to_f64
+body: |
+  bb.0:
+    liveins: $vgpr0
+
+    ; CHECK-LABEL: name: test_fpext_f16_to_f64
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+    ; CHECK: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
+    ; CHECK: [[FPEXT1:%[0-9]+]]:_(s64) = G_FPEXT [[FPEXT]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[FPEXT1]](s64)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s16) = G_TRUNC %0
+    %2:_(s64) = G_FPEXT %1
+    $vgpr0_vgpr1 = COPY %2
+...
+
+---
+name: test_fpext_v2f16_to_v2f64
+body: |
+  bb.0:
+    liveins: $vgpr0
+
+    ; CHECK-LABEL: name: test_fpext_v2f16_to_v2f64
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; CHECK: [[FPEXT:%[0-9]+]]:_(s32) = nnan G_FPEXT [[TRUNC]](s16)
+    ; CHECK: [[FPEXT1:%[0-9]+]]:_(s64) = G_FPEXT [[FPEXT]](s32)
+    ; CHECK: [[FPEXT2:%[0-9]+]]:_(s32) = nnan G_FPEXT [[TRUNC1]](s16)
+    ; CHECK: [[FPEXT3:%[0-9]+]]:_(s64) = G_FPEXT [[FPEXT2]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FPEXT1]](s64), [[FPEXT3]](s64)
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s64>) = nnan G_FPEXT %0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
+...