[NFC][Driver][HIP] Fix mixing amdgcnspirv and gfxXXX via --offload-arch

[AlexVlx]

Due to amdgcnspirv piggybacking on the HIPAMDToolchain, it loses its immediately apparent SPIR-Vness, which makes the Driver want to go all the way to Assembly emmission. This was problematic as we were trying to llvm-link SPIR-V, before trying to translate it. This patch ensures that we do the right thing and stop at bitcode emission if we are mixing amdgcnspirv and concrete targets.

[llvmbot]

@llvm/pr-subscribers-clang

Author: Alex Voicu (AlexVlx)

Changes

Due to amdgcnspirv piggybacking on the HIPAMDToolchain, it loses its immediately apparent SPIR-Vness, which makes the Driver want to go all the way to Assembly emmission. This was problematic as we were trying to llvm-link SPIR-V, before trying to translate it. This patch ensures that we do the right thing and stop at bitcode emission if we are mixing amdgcnspirv and concrete targets.

---

Full diff: https://github.com/llvm/llvm-project/pull/133024.diff

1 Files Affected:

• (modified) clang/lib/Driver/Driver.cpp (+5-2)

diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index 056bfcf1b739a..07e36ea2efba4 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -3750,9 +3750,12 @@ class OffloadingActionBuilder final {
             // compiler phases, including backend and assemble phases.
             ActionList AL;
             Action *BackendAction = nullptr;
-            if (ToolChains.front()->getTriple().isSPIRV()) {
+            if (ToolChains.front()->getTriple().isSPIRV() ||
+                (ToolChains.front()->getTriple().isAMDGCN() &&
+                 GpuArchList[I] == StringRef("amdgcnspirv"))) {
               // Emit LLVM bitcode for SPIR-V targets. SPIR-V device tool chain
-              // (HIPSPVToolChain) runs post-link LLVM IR passes.
+              // (HIPSPVToolChain or HIPAMDToolChain) runs post-link LLVM IR
+              // passes.
               types::ID Output = Args.hasArg(options::OPT_S)
                                      ? types::TY_LLVM_IR
                                      : types::TY_LLVM_BC;

[llvmbot]

@llvm/pr-subscribers-clang-driver

Author: Alex Voicu (AlexVlx)

Changes

Due to amdgcnspirv piggybacking on the HIPAMDToolchain, it loses its immediately apparent SPIR-Vness, which makes the Driver want to go all the way to Assembly emmission. This was problematic as we were trying to llvm-link SPIR-V, before trying to translate it. This patch ensures that we do the right thing and stop at bitcode emission if we are mixing amdgcnspirv and concrete targets.

---

Full diff: https://github.com/llvm/llvm-project/pull/133024.diff

1 Files Affected:

• (modified) clang/lib/Driver/Driver.cpp (+5-2)

diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index 056bfcf1b739a..07e36ea2efba4 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -3750,9 +3750,12 @@ class OffloadingActionBuilder final {
             // compiler phases, including backend and assemble phases.
             ActionList AL;
             Action *BackendAction = nullptr;
-            if (ToolChains.front()->getTriple().isSPIRV()) {
+            if (ToolChains.front()->getTriple().isSPIRV() ||
+                (ToolChains.front()->getTriple().isAMDGCN() &&
+                 GpuArchList[I] == StringRef("amdgcnspirv"))) {
               // Emit LLVM bitcode for SPIR-V targets. SPIR-V device tool chain
-              // (HIPSPVToolChain) runs post-link LLVM IR passes.
+              // (HIPSPVToolChain or HIPAMDToolChain) runs post-link LLVM IR
+              // passes.
               types::ID Output = Args.hasArg(options::OPT_S)
                                      ? types::TY_LLVM_IR
                                      : types::TY_LLVM_BC;

[jhuber6]

Seems in line with the weird workarounds we already use.