[AMDGPU] Fix image opcodes GlobalISel on gfx90a.

[rampitec]

Summary

• Correct flavor of an instruction was not selected.
• GFX90A does not support TFE.

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

1648

Probably should just return false and turn the error message into a debug print

[rampitec]

ping

[Joe_Nash]

LGTM

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

1716

Same applies here? Return false instead of error

[rampitec]

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

1716

D126797

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

bool AMDGPUInstructionSelector::selectImageIntrinsic(
 
   int Opcode = -1;
   if (IsGFX10Plus) {
     Opcode = AMDGPU::getMIMGOpcode(IntrOpcode,
                                    UseNSA ? AMDGPU::MIMGEncGfx10NSA
                                           : AMDGPU::MIMGEncGfx10Default,
                                    NumVDataDwords, NumVAddrDwords);
   } else {
-    if (STI.getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS)
+    if (Subtarget->hasGFX90AInsts()) {
+      Opcode = AMDGPU::getMIMGOpcode(IntrOpcode, AMDGPU::MIMGEncGfx90a,
+                                     NumVDataDwords, NumVAddrDwords);
+      if (Opcode == -1) {
+        LLVM_DEBUG(
+            dbgs()
+            << "requested image instruction is not supported on this GPU\n");
+        return false;
+      }
+    }
+    if (Opcode == -1 &&
+        STI.getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS)
       Opcode = AMDGPU::getMIMGOpcode(IntrOpcode, AMDGPU::MIMGEncGfx8,
                                      NumVDataDwords, NumVAddrDwords);
     if (Opcode == -1)
       Opcode = AMDGPU::getMIMGOpcode(IntrOpcode, AMDGPU::MIMGEncGfx6,
                                      NumVDataDwords, NumVAddrDwords);
   }
   assert(Opcode != -1);
 
bool AMDGPUInstructionSelector::selectImageIntrinsic(
   MIB.addImm(Unorm);
 
   MIB.addImm(CPol);
   MIB.addImm(IsA16 &&  // a16 or r128
              STI.hasFeature(AMDGPU::FeatureR128A16) ? -1 : 0);
   if (IsGFX10Plus)
     MIB.addImm(IsA16 ? -1 : 0);
 
+  if (!Subtarget->hasGFX90AInsts())
     MIB.addImm(TFE); // tfe
+  else if (TFE)
+    report_fatal_error("TFE is not supported on this GPU");
+
   MIB.addImm(LWE); // lwe
   if (!IsGFX10Plus)
     MIB.addImm(DimInfo->DA ? -1 : 0);
   if (BaseOpcode->HasD16)
     MIB.addImm(IsD16 ? -1 : 0);
 
   if (IsTexFail) {
     // An image load instruction with TFE/LWE only conditionally writes to its

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.image.atomic.dim.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -o - %s | FileCheck -check-prefix=GFX6 %s
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -o - %s | FileCheck -check-prefix=GFX8 %s
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -o - %s | FileCheck -check-prefix=GFX900 %s
-; XUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx90a -o - %s | FileCheck -check-prefix=GFX90A %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx90a -o - %s | FileCheck -check-prefix=GFX90A %s
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -o - %s | FileCheck -check-prefix=GFX10 %s
 
-; FIXME: global isel does not select _gfx90a flavor of image instructions.
-
 define amdgpu_ps float @atomic_swap_i32_1d(<8 x i32> inreg %rsrc, i32 %data, i32 %s) {
 ; GFX6-LABEL: atomic_swap_i32_1d:
 ; GFX6:       ; %bb.0: ; %main_body
 ; GFX6-NEXT:    s_mov_b32 s0, s2
 ; GFX6-NEXT:    s_mov_b32 s1, s3
 ; GFX6-NEXT:    s_mov_b32 s2, s4
 ; GFX6-NEXT:    s_mov_b32 s3, s5
 ; GFX6-NEXT:    s_mov_b32 s4, s6

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.dim.gfx90a.ll

 ; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s
-; XUN: llc -global-isel -march=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s
-
-; FIXME: global isel does not select _gfx90a flavor of image instructions.
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s
 
 ; GCN-LABEL: {{^}}load_1d:
 ; GCN: image_load v[0:3], v0, s[0:7] dmask:0xf unorm{{$}}
 define amdgpu_ps <4 x float> @load_1d(<8 x i32> inreg %rsrc, i32 %s) {
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 0)
   ret <4 x float> %v
 }

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.dim.gfx90a.ll

 ; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX90A,SDAG %s
-; XUN: llc -global-isel -march=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX90A,GISEL %s
-
-; FIXME: global isel does not select _gfx90a flavor of image instructions.
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX90A,GISEL %s
 
 ; GFX90A-LABEL: {{^}}sample_1d:
 ; GFX90A-NOT: s_wqm_b64
 ; GFX90A:     image_sample v[{{[0-9:]+}}], v{{[0-9]+}}, s[{{[0-9:]+}}], s[{{[0-9:]+}}] dmask:0xf
 define amdgpu_ps <4 x float> @sample_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
   ret <4 x float> %v