[PATCH] D25589: AMDGPU: Fix using incorrect private resource with no allocation

[Tom Stellard via llvm-commits]

tstellarAMD added inline comments.


================
Comment at: lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp:465-471
+    ExtraSGPRs += 2;
 
-  if (STM.getGeneration() < SISubtarget::VOLCANIC_ISLANDS) {
-    if (FlatUsed)
-      ExtraSGPRs = 4;
-  } else {
-    if (STM.isXNACKEnabled())
-      ExtraSGPRs = 4;
+  if (FlatUsed)
+    ExtraSGPRs += 2;
 
-    if (FlatUsed)
-      ExtraSGPRs = 6;
-  }
+  if (STM.isXNACKEnabled())
+    ExtraSGPRs += 2;
----------------
I don't think this is correct.  For example if the shader uses 12 SGPRS , VCC and  Flat.  then we would allocate 12 + 4 = 16 SGPRs.  This would put the flat_scratch register at s[10:11], which may also be referenced as a regular SGPR by the shader, so we may end up randomly over writing this register.


https://reviews.llvm.org/D25589