Skip to content

Commit 5d2902d

Browse files
Add comment
1 parent 328e12c commit 5d2902d

File tree

1 file changed

+7
-0
lines changed

1 file changed

+7
-0
lines changed

llvm/lib/Target/NVPTX/NVPTXSubtarget.h

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -95,6 +95,13 @@ class NVPTXSubtarget : public NVPTXGenSubtargetInfo {
9595
bool hasDotInstructions() const {
9696
return SmVersion >= 61 && PTXVersion >= 50;
9797
}
98+
// Prior to CUDA 12.3 ptxas did not recognize that the trap instruction
99+
// terminates a basic block. Instead, it would assume that control flow
100+
// continued to the next instruction. The next instruction could be in the
101+
// block that's lexically below it. This would lead to a phantom CFG edges
102+
// being created within ptxas. This issue was fixed in CUDA 12.3. Thus, when
103+
// PTX ISA versions 8.3+ we can confidently say that the bug will not be
104+
// present.
98105
bool hasPTXASUnreachableBug() const { return PTXVersion < 83; }
99106
bool hasCvtaParam() const { return SmVersion >= 70 && PTXVersion >= 77; }
100107
unsigned int getFullSmVersion() const { return FullSmVersion; }

0 commit comments

Comments
 (0)