[flang-commits] [flang] [llvm] Fixing the location attribute added to mapInfoOp (PR #90764)
Anchu Rajendran S via flang-commits
flang-commits at lists.llvm.org
Wed May 22 23:53:01 PDT 2024
https://github.com/anchuraj updated https://github.com/llvm/llvm-project/pull/90764
>From cf8ebf7a9e86e65d74057bc168e8ea0533538032 Mon Sep 17 00:00:00 2001
From: Anchu Rajendran <asudhaku at amd.com>
Date: Wed, 1 May 2024 13:22:23 -0500
Subject: [PATCH] Fixing the location attribute added to mapInfoOp
Named location attribute added tgt_info shall be used by runtime calls
like ompx_dump_mapping_tables to print the information of variables that
are mapped to the device. Before this change, ompx_dump_mapping_tables
was printing the wrong location information.
---
flang/lib/Lower/OpenMP/ClauseProcessor.cpp | 5 ++-
flang/lib/Lower/OpenMP/OpenMP.cpp | 8 +++-
.../offloading/fortran/dump_map_tables.f90 | 38 +++++++++++++++++++
3 files changed, 48 insertions(+), 3 deletions(-)
create mode 100644 offload/test/offloading/fortran/dump_map_tables.f90
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index 875599098b3dc..68619f699ebb2 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -882,8 +882,11 @@ bool ClauseProcessor::processMap(
// Explicit map captures are captured ByRef by default,
// optimisation passes may alter this to ByCopy or other capture
// types to optimise
+ auto location = mlir::NameLoc::get(
+ mlir::StringAttr::get(firOpBuilder.getContext(), asFortran.str()),
+ symAddr.getLoc());
mlir::omp::MapInfoOp mapOp = createMapInfoOp(
- firOpBuilder, clauseLocation, symAddr,
+ firOpBuilder, location, symAddr,
/*varPtrPtr=*/mlir::Value{}, asFortran.str(), bounds,
/*members=*/{}, /*membersIndex=*/mlir::DenseIntElementsAttr{},
static_cast<
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 17b362cc2f32e..e57752424994b 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -1604,9 +1604,13 @@ genTargetOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
mapFlag |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_TO;
mapFlag |= llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_FROM;
}
-
+ auto context = converter.getFirOpBuilder().getContext();
+ auto location =
+ mlir::NameLoc::get(mlir::StringAttr::get(firOpBuilder.getContext(),
+ sym.name().ToString()),
+ baseOp.getLoc());
mlir::Value mapOp = createMapInfoOp(
- firOpBuilder, baseOp.getLoc(), baseOp, /*varPtrPtr=*/mlir::Value{},
+ firOpBuilder, location, baseOp, /*varPtrPtr=*/mlir::Value{},
name.str(), bounds, /*members=*/{},
/*membersIndex=*/mlir::DenseIntElementsAttr{},
static_cast<
diff --git a/offload/test/offloading/fortran/dump_map_tables.f90 b/offload/test/offloading/fortran/dump_map_tables.f90
new file mode 100644
index 0000000000000..cb66ef348e3c2
--- /dev/null
+++ b/offload/test/offloading/fortran/dump_map_tables.f90
@@ -0,0 +1,38 @@
+! Offloading test with runtine call to ompx_dump_mapping_tables
+! Fortran array writing some values and printing the variable mapped to device
+! correctly receives the updates made on the device.
+! REQUIRES: flang
+! UNSUPPORTED: nvptx64-nvidia-cuda-LTO
+! UNSUPPORTED: aarch64-unknown-linux-gnu
+! UNSUPPORTED: aarch64-unknown-linux-gnu-LTO
+! UNSUPPORTED: x86_64-pc-linux-gnu
+! UNSUPPORTED: x86_64-pc-linux-gnu-LTO
+
+! RUN: %libomptarget-compile-fortran-run-and-check-generic
+
+program map_dump_example
+ INTERFACE
+ SUBROUTINE ompx_dump_mapping_tables() BIND(C)
+ END SUBROUTINE ompx_dump_mapping_tables
+ END INTERFACE
+
+ integer i,j,k,N
+ integer async_q(4)
+ real :: A(5000000)
+ N=5000000
+ do i=1, N
+ A(i)=0
+ enddo
+! clang-format off
+! CHECK: omptarget device 0 info: OpenMP Host-Device pointer mappings after block
+! CHECK-NEXT: omptarget device 0 info: Host Ptr Target Ptr Size (B) DynRefCount HoldRefCount Declaration
+! CHECK-NEXT: omptarget device 0 info: {{(0x[0-9a-f]{16})}} {{(0x[0-9a-f]{16})}} 20000000 1 0 {{.*}} at a(:n):21:11
+! clang-format on
+!$omp target enter data map(to:A(:N))
+ call ompx_dump_mapping_tables()
+!$omp target parallel do
+ do i=1, N
+ A(i)=A(i)*2
+ enddo
+!$omp target exit data map(from:A)
+end program
More information about the flang-commits
mailing list