This revision was automatically updated to reflect the committed changes.
Closed by commit rL270150: [CUDA] Implement __ldg using intrinsics. (authored
by jlebar).
Changed prior to commit:
http://reviews.llvm.org/D19990?vs=56603&id=57873#toc
Repository:
rL LLVM
http://reviews.llvm.org/D1999
rsmith accepted this revision.
rsmith added a comment.
This revision is now accepted and ready to land.
After offline discussion: we don't know for sure whether we're going to hit the
combinatorial explosion in future or not. Let's go ahead with this as-is for
now, then, with the explicit acknow
rsmith added inline comments.
Comment at: include/clang/Basic/BuiltinsNVPTX.def:569-603
@@ -568,1 +568,37 @@
+// __ldg. This is not implemented as a builtin by nvcc.
+BUILTIN(__nvvm_ldg_c, "ccC*", "")
+BUILTIN(__nvvm_ldg_s, "ssC*", "")
+BUILTIN(__nvvm_ldg_i, "iiC*", "")
+BUILTI
jlebar added a comment.
Friendly ping. This is a big help with some Tensorflow benchmarks.
http://reviews.llvm.org/D19990
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
jlebar updated this revision to Diff 56603.
jlebar added a comment.
Remove static_asserts.
http://reviews.llvm.org/D19990
Files:
include/clang/Basic/BuiltinsNVPTX.def
lib/CodeGen/CGBuiltin.cpp
lib/Headers/CMakeLists.txt
lib/Headers/__clang_cuda_intrinsics.h
lib/Headers/__clang_cuda_ru
jlebar added a comment.
Art pointed out that static_assert is c++11-only.
I'll just remove them and make a note to move them into the CUDA test-suite
stuff Art is working on.
http://reviews.llvm.org/D19990
___
cfe-commits mailing list
cfe-commits@
tra added a comment.
OK. Let's stick with __ldg for now.
http://reviews.llvm.org/D19990
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
jlebar added a comment.
Art pointed me to the fact that CUDA 8 adds a bunch more load intrinsics, and I
said ohmygosh maybe we *do* want to do the variadic intrinsic thing here.
But now looking at how __builtin_add_overflow is implemented, we'd need special
sema checking to make it work. We wo
jlebar added inline comments.
Comment at: include/clang/Basic/BuiltinsNVPTX.def:569-603
@@ -568,1 +568,37 @@
+// __ldg. This is not implemented as a builtin by nvcc.
+BUILTIN(__nvvm_ldg_c, "ccC*", "")
+BUILTIN(__nvvm_ldg_s, "ssC*", "")
+BUILTIN(__nvvm_ldg_i, "iiC*", "")
+BUILTI
jlebar added inline comments.
Comment at: include/clang/Basic/BuiltinsNVPTX.def:569-603
@@ -568,1 +568,37 @@
+// __ldg. This is not implemented as a builtin by nvcc.
+BUILTIN(__nvvm_ldg_c, "ccC*", "")
+BUILTIN(__nvvm_ldg_s, "ssC*", "")
+BUILTIN(__nvvm_ldg_i, "iiC*", "")
+BUILTI
majnemer added a subscriber: majnemer.
Comment at: include/clang/Basic/BuiltinsNVPTX.def:569-603
@@ -568,1 +568,37 @@
+// __ldg. This is not implemented as a builtin by nvcc.
+BUILTIN(__nvvm_ldg_c, "ccC*", "")
+BUILTIN(__nvvm_ldg_s, "ssC*", "")
+BUILTIN(__nvvm_ldg_i, "iiC*", ""
11 matches
Mail list logo