summaryrefslogtreecommitdiffstats
path: root/source/slang/slang-emit.cpp
diff options
context:
space:
mode:
authorArielG-NV <159081215+ArielG-NV@users.noreply.github.com>2025-08-07 00:22:22 -0700
committerGitHub <noreply@github.com>2025-08-07 07:22:22 +0000
commit063cbeaaea2fb00a10c6058ea4a9632092772ea5 (patch)
treeb4412347d6c264c3b1a84ec971921a5e2fe76134 /source/slang/slang-emit.cpp
parent9e2685853033f4286feaf22d04a755a7395d95ce (diff)
Initial copy elision pass (#8042)
Fixes #7574 Changes: * Add an initial (fairly simple) optimization pass which is able to eliminate redundant copies. * Our current existing optimizer passes remove redundant load/store very robustly, this pass will focus on other cases of copy elimination * Primary approach is to make all functions which are `in T` and `T` is trivial to copy into a `__constref T`. We then (depending on scenario) manually insert a variable+load if a pass-by-reference is not possible; otherwise we pass by `constref`. * Added optimizations to eliminate redundant code which causes `constref` to fail to compile --------- Co-authored-by: Harsh Aggarwal <haaggarwal@nvidia.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: slangbot <ellieh+slangbot@nvidia.com> Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>
Diffstat (limited to 'source/slang/slang-emit.cpp')
-rw-r--r--source/slang/slang-emit.cpp7
1 files changed, 7 insertions, 0 deletions
diff --git a/source/slang/slang-emit.cpp b/source/slang/slang-emit.cpp
index 405bca5a2..7d8f1438d 100644
--- a/source/slang/slang-emit.cpp
+++ b/source/slang/slang-emit.cpp
@@ -110,6 +110,7 @@
#include "slang-ir-strip-default-construct.h"
#include "slang-ir-strip-legalization-insts.h"
#include "slang-ir-synthesize-active-mask.h"
+#include "slang-ir-transform-params-to-constref.h"
#include "slang-ir-translate-global-varying-var.h"
#include "slang-ir-undo-param-copy.h"
#include "slang-ir-uniformity.h"
@@ -1714,6 +1715,12 @@ Result linkAndOptimizeIR(
// For CUDA/OptiX like targets, add our pass to replace inout parameter copies with direct
// pointers
undoParameterCopy(irModule);
+ // Transform struct parameters to use ConstRef for better performance
+ if (isCPUTarget(targetRequest) || isCUDATarget(targetRequest) ||
+ isMetalTarget(targetRequest))
+ {
+ transformParamsToConstRef(irModule, codeGenContext->getSink());
+ }
#if 0
dumpIRIfEnabled(codeGenContext, irModule, "PARAMETER COPIES REPLACED WITH DIRECT POINTERS");
#endif