[clang] [llvm] [clang][SPIR-V] Add support for AMDGCN flavoured SPIRV (PR #89796)

[Alex Voicu via llvm-commits]

================
@@ -309,7 +309,45 @@ StringRef AMDGPU::getCanonicalArchName(const Triple &T, StringRef Arch) {
 void AMDGPU::fillAMDGPUFeatureMap(StringRef GPU, const Triple &T,
                                   StringMap<bool> &Features) {
   // XXX - What does the member GPU mean if device name string passed here?
-  if (T.isAMDGCN()) {
+  if (T.isSPIRV() && T.getOS() == Triple::OSType::AMDHSA) {
+    // AMDGCN SPIRV must support the union of all AMDGCN features.
+    Features["atomic-ds-pk-add-16-insts"] = true;
+    Features["atomic-flat-pk-add-16-insts"] = true;
+    Features["atomic-buffer-global-pk-add-f16-insts"] = true;
+    Features["atomic-global-pk-add-bf16-inst"] = true;
+    Features["atomic-fadd-rtn-insts"] = true;
+    Features["ci-insts"] = true;
+    Features["dot1-insts"] = true;
+    Features["dot2-insts"] = true;
+    Features["dot3-insts"] = true;
+    Features["dot4-insts"] = true;
+    Features["dot5-insts"] = true;
+    Features["dot7-insts"] = true;
+    Features["dot8-insts"] = true;
+    Features["dot9-insts"] = true;
+    Features["dot10-insts"] = true;
+    Features["dot11-insts"] = true;
+    Features["dl-insts"] = true;
+    Features["16-bit-insts"] = true;
+    Features["dpp"] = true;
+    Features["gfx8-insts"] = true;
+    Features["gfx9-insts"] = true;
+    Features["gfx90a-insts"] = true;
+    Features["gfx940-insts"] = true;
+    Features["gfx10-insts"] = true;
+    Features["gfx10-3-insts"] = true;
+    Features["gfx11-insts"] = true;
+    Features["gfx12-insts"] = true;
----------------
AlexVlx wrote:

One can neither say it's available nor say it's not, it's undecidable at SPIRV generation time (if we knew, we'd not need it in the first place and would've just targeted the actual hardware). The idea is that AMDGCN SPIRV will "support" the union of AMDGCN features, so that the associated builtins / constructs are retained in the generated SPIRV. Afterwards, at JIT time, one handles matching to the actual physical target that is present; this is done via a piece of paired functionality in the ME which has not yet been submitted.

https://github.com/llvm/llvm-project/pull/89796