Fix OptiX data layout mismatch with LLVM 16+

i128:128 was already in the NVPTX backend since LLVM 6, the docs this was taken from seem outdated. Unclear it started failing now, but the data layout seems clearly incomplete without this. Signed-off-by: Brecht Van Lommel <[email protected]>
AcademySoftwareFoundation · Sep 28, 2023 · dee0540 · dee0540
1 parent c8e684e
commit dee0540
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/src/liboslexec/llvm_instance.cpp b/src/liboslexec/llvm_instance.cpp
@@ -1748,7 +1748,7 @@ BackendLLVM::run()
             // The target triple and data layout used here are those specified
             // for NVPTX (https://www.llvm.org/docs/NVPTXUsage.html#triples).
             ll.module()->setDataLayout(
-                "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64");
+                "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-i128:128:128-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64");
             ll.module()->setTargetTriple("nvptx64-nvidia-cuda");
         }
 #    endif
@@ -1813,7 +1813,7 @@ BackendLLVM::run()
                     err);
 
             shadeops_module->setDataLayout(
-                "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64");
+                "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-i128:128:128-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64");
             shadeops_module->setTargetTriple("nvptx64-nvidia-cuda");
 
             std::unique_ptr<llvm::Module> shadeops_ptr(shadeops_module);
@@ -1840,7 +1840,7 @@ BackendLLVM::run()
                         err);
 
                 rend_lib_module->setDataLayout(
-                    "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64");
+                    "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-i128:128:128-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64");
                 rend_lib_module->setTargetTriple("nvptx64-nvidia-cuda");
 
                 for (llvm::Function& fn : *rend_lib_module) {
@@ -1860,7 +1860,7 @@ BackendLLVM::run()
             // See: https://llvm.org/docs/NVPTXUsage.html
             ll.module()->setTargetTriple("nvptx64-nvidia-cuda");
             ll.module()->setDataLayout(
-                "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64");
+                "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-i128:128:128-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64");
 
             // Tag each function as an OSL library function to help with
             // inlining and optimization after codegen.