AMDGPU: Avoid assertions after SGPR spilling failed

Summary:
The comment explains it: emitError does not necessarily exit the compilation
process, and then using NoRegister leads to assertions later on.
This generates incorrect code, of course, but the user should know to not use
the result when an error has been emitted.

It would be nice to have a test-case for this inside the LLVM repository,
but llc exits on error. shader-db tests trigger the underlying issue at least
on Tonga.

Reviewers: arsenm, tstellarAMD, mareko

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D15826

llvm-svn: 256757

1 files changed, 11 insertions, 0 deletions

diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
index 935aad42719..bf15516bea7 100644
--- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
@@ -156,6 +156,17 @@ SIMachineFunctionInfo::SpilledReg SIMachineFunctionInfo::getSpilledReg(
 
   if (!LaneVGPRs.count(LaneVGPRIdx)) {
     unsigned LaneVGPR = TRI->findUnusedRegister(MRI, &AMDGPU::VGPR_32RegClass);
+
+    if (LaneVGPR == AMDGPU::NoRegister) {
+      LLVMContext &Ctx = MF->getFunction()->getContext();
+      Ctx.emitError("Ran out of VGPRs for spilling SGPR");
+
+      // When compiling from inside Mesa, the compilation continues.
+      // Select an arbitrary register to avoid triggering assertions
+      // during subsequent passes.
+      LaneVGPR = AMDGPU::VGPR0;
+    }
+
     LaneVGPRs[LaneVGPRIdx] = LaneVGPR;
 
     // Add this register as live-in to all blocks to avoid machine verifer