AMDGPU: Avoid report_fatal_error in image intrinsic lowering

[arsenm]

No description provided.

[arsenm]

• AMDGPU: Avoid report_fatal_error in image intrinsic lowering #145201 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Matt Arsenault (arsenm)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/145201.diff

2 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/SIISelLowering.cpp (+24-6)
• (modified) llvm/test/CodeGen/AMDGPU/unsupported-image-sample.ll (+14-5)

diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index b9023b6d7a3a6..0efbf381c67f7 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -8668,11 +8668,15 @@ SDValue SITargetLowering::lowerImage(SDValue Op,
                     : False);
   if (IsGFX10Plus)
     Ops.push_back(IsA16 ? True : False);
-  if (!Subtarget->hasGFX90AInsts()) {
+
+  if (!Subtarget->hasGFX90AInsts())
     Ops.push_back(TFE); // tfe
-  } else if (TFE->getAsZExtVal()) {
-    report_fatal_error("TFE is not supported on this GPU");
+  else if (TFE->getAsZExtVal()) {
+    DAG.getContext()->diagnose(DiagnosticInfoUnsupported(
+        DAG.getMachineFunction().getFunction(),
+        "TFE is not supported on this GPU", DL.getDebugLoc()));
   }
+
   if (!IsGFX12Plus || BaseOpcode->Sampler || BaseOpcode->MSAA)
     Ops.push_back(LWE); // lwe
   if (!IsGFX10Plus)
@@ -8703,9 +8707,23 @@ SDValue SITargetLowering::lowerImage(SDValue Op,
     if (Subtarget->hasGFX90AInsts()) {
       Opcode = AMDGPU::getMIMGOpcode(IntrOpcode, AMDGPU::MIMGEncGfx90a,
                                      NumVDataDwords, NumVAddrDwords);
-      if (Opcode == -1)
-        report_fatal_error(
-            "requested image instruction is not supported on this GPU");
+      if (Opcode == -1) {
+        DAG.getContext()->diagnose(DiagnosticInfoUnsupported(
+            DAG.getMachineFunction().getFunction(),
+            "requested image instruction is not supported on this GPU",
+            DL.getDebugLoc()));
+
+        unsigned Idx = 0;
+        SmallVector<SDValue, 3> RetValues(OrigResultTypes.size());
+        for (EVT VT : OrigResultTypes) {
+          if (VT == MVT::Other)
+            RetValues[Idx++] = Op.getOperand(0); // Chain
+          else
+            RetValues[Idx++] = DAG.getPOISON(VT);
+        }
+
+        return DAG.getMergeValues(RetValues, DL);
+      }
     }
     if (Opcode == -1 &&
         Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS)
diff --git a/llvm/test/CodeGen/AMDGPU/unsupported-image-sample.ll b/llvm/test/CodeGen/AMDGPU/unsupported-image-sample.ll
index db5f0ad42a677..b3cf3790a59d3 100644
--- a/llvm/test/CodeGen/AMDGPU/unsupported-image-sample.ll
+++ b/llvm/test/CodeGen/AMDGPU/unsupported-image-sample.ll
@@ -1,17 +1,16 @@
 ; RUN: llc -O0 -mtriple=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
 ; RUN: llc -O0 -mtriple=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
 ; RUN: llc -O0 -mtriple=amdgcn -mcpu=gfx9-generic --amdhsa-code-object-version=6 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
-; RUN: not --crash llc -O0 -mtriple=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s 2>&1 | FileCheck -check-prefixes=GFX90A %s
-; RUN: not --crash llc -O0 -mtriple=amdgcn -mcpu=gfx942 -verify-machineinstrs < %s 2>&1 | FileCheck -check-prefixes=GFX942 %s
+; RUN: not llc -O0 -mtriple=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s 2>&1 | FileCheck -check-prefixes=GFX90A %s
+; RUN: not llc -O0 -mtriple=amdgcn -mcpu=gfx942 -verify-machineinstrs < %s 2>&1 | FileCheck -check-prefixes=GFX942 %s
 ; RUN: llc -O0 -mtriple=amdgcn -mcpu=gfx1030 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX1030 %s
 ; RUN: llc -O0 -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX1100 %s
 
 ; GFX9-LABEL: image_sample_test:
 ; GFX9: image_sample_lz
 
-; GFX90A: LLVM ERROR: requested image instruction is not supported on this GPU
-
-; GFX942: LLVM ERROR: requested image instruction is not supported on this GPU
+; GFX90A: error: <unknown>:0:0: in function image_sample_test void (ptr addrspace(1), float, float, <8 x i32>, <4 x i32>): requested image instruction is not supported on this GPU
+; GFX942: error: <unknown>:0:0: in function image_sample_test void (ptr addrspace(1), float, float, <8 x i32>, <4 x i32>): requested image instruction is not supported on this GPU
 
 ; GFX1030-LABEL: image_sample_test:
 ; GFX1030: image_sample_lz
@@ -28,3 +27,13 @@ define amdgpu_kernel void @image_sample_test(ptr addrspace(1) %out, float %arg1,
 }
 
 declare <4 x float> @llvm.amdgcn.image.sample.lz.2d.v4f32.f32(i32 immarg, float, float, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg)
+
+; GFX90A: error: <unknown>:0:0: in function sample_1d_tfe <4 x float> (<8 x i32>, <4 x i32>, ptr addrspace(1), float): TFE is not supported on this GPU
+; GFX942: error: <unknown>:0:0: in function sample_1d_tfe <4 x float> (<8 x i32>, <4 x i32>, ptr addrspace(1), float): TFE is not supported on this GPU
+define <4 x float> @sample_1d_tfe(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, ptr addrspace(1) inreg %out, float %s) {
+  %v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
+  %v.vec = extractvalue {<4 x float>, i32} %v, 0
+  %v.err = extractvalue {<4 x float>, i32} %v, 1
+  store i32 %v.err, ptr addrspace(1) %out, align 4
+  ret <4 x float> %v.vec
+}