[Mesa-dev] [PATCH] glsl: always call do_lower_jumps() after loop unrolling
Timothy Arceri
tarceri at itsqueeze.com
Tue Apr 3 01:38:13 UTC 2018
This fixes a bug in radeonsi where LLVM cannot handle the case where
a break exists but its not the last instruction in the block.
LLVM would fail with:
Terminator found in the middle of a basic block!
LLVM ERROR: Broken function found, compilation aborted!
Fixes: 96fe8834f539 "glsl_to_tgsi: do fewer optimizations with GLSLOptimizeConservatively"
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105317
---
src/compiler/glsl/glsl_parser_extras.cpp | 18 ++++++++++++++++++
1 file changed, 18 insertions(+)
diff --git a/src/compiler/glsl/glsl_parser_extras.cpp b/src/compiler/glsl/glsl_parser_extras.cpp
index 275c4d75711..0cc57f5a887 100644
--- a/src/compiler/glsl/glsl_parser_extras.cpp
+++ b/src/compiler/glsl/glsl_parser_extras.cpp
@@ -2242,6 +2242,24 @@ do_common_optimization(exec_list *ir, bool linked,
loop_progress = false;
loop_progress |= do_constant_propagation(ir);
loop_progress |= do_if_simplification(ir);
+
+ /* Some drivers only call do_common_optimization() once rather
+ * than in a loop. So we must call do_lower_jumps() after
+ * unrolling a loop because for drivers that use LLVM validation
+ * will fail if a jump is not the last instruction in the block.
+ * For example the following will fail LLVM validation:
+ *
+ * (loop (
+ * ...
+ * break
+ * (assign (x) (var_ref v124) (expression int + (var_ref v124)
+ * (constant int (1)) ) )
+ * ))
+ */
+ loop_progress |= do_lower_jumps(ir, true, true,
+ options->EmitNoMainReturn,
+ options->EmitNoCont,
+ options->EmitNoLoops);
}
progress |= loop_progress;
}
--
2.14.3
More information about the mesa-dev
mailing list