maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/OpenMP/nvptx_data_sharing.cpp

58 lines
2.3 KiB
C++
Raw Normal View History

[OpenMP] Add implicit data sharing support when offloading to NVIDIA GPUs using OpenMP device offloading Summary: This patch is part of the development effort to add support in the current OpenMP GPU offloading implementation for implicitly sharing variables between a target region executed by the team master thread and the worker threads within that team. This patch is the first of three required for successfully performing the implicit sharing of master thread variables with the worker threads within a team. The remaining two patches are: - Patch D38978 to the LLVM NVPTX backend which ensures the lowering of shared variables to an device memory which allows the sharing of references; - Patch (coming soon) is a patch to libomptarget runtime library which ensures that a list of references to shared variables is properly maintained. A simple code snippet which illustrates an implicit data sharing situation is as follows: ``` #pragma omp target { // master thread only int v; #pragma omp parallel { // worker threads // use v } } ``` Variable v is implicitly shared from the team master thread which executes the code in between the target and parallel directives. The worker threads must operate on the latest version of v, including any updates performed by the master. The code generated in this patch relies on the LLVM NVPTX patch (mentioned above) which prevents v from being lowered in the thread local memory of the master thread thus making the reference to this variable un-shareable with the workers. This ensures that the code generated by this patch is correct. Since the parallel region is outlined the passing of arguments to the outlined regions must preserve the original order of arguments. The runtime therefore maintains a list of references to shared variables thus ensuring their passing in the correct order. The passing of arguments to the outlined parallel function is performed in a separate function which the data sharing infrastructure constructs in this patch. The function is inlined when optimizations are enabled. Reviewers: hfinkel, carlo.bertolli, arpith-jacob, Hahnfeld, ABataev, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D38976 llvm-svn: 318773
2017-11-21 23:54:54 +08:00
// Test device data sharing codegen.
///==========================================================================///
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CK1
// expected-no-diagnostics
#ifndef HEADER
#define HEADER
void test_ds(){
#pragma omp target
{
int a = 10;
#pragma omp parallel
{
a = 1000;
}
}
}
/// ========= In the worker function ========= ///
[OpenMP] Add function attribute for triggering data sharing. Summary: The backend should only emit data sharing code for the cases where it is needed. A new function attribute is used by Clang to enable data sharing only for the cases where OpenMP semantics require it and there are variables that need to be shared. Reviewers: hfinkel, Hahnfeld, ABataev, carlo.bertolli, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D41123 llvm-svn: 320527
2017-12-13 05:38:43 +08:00
// CK1: define internal void @__omp_offloading_{{.*}}test_ds{{.*}}worker() [[ATTR1:#.*]] {
[OpenMP] Add implicit data sharing support when offloading to NVIDIA GPUs using OpenMP device offloading Summary: This patch is part of the development effort to add support in the current OpenMP GPU offloading implementation for implicitly sharing variables between a target region executed by the team master thread and the worker threads within that team. This patch is the first of three required for successfully performing the implicit sharing of master thread variables with the worker threads within a team. The remaining two patches are: - Patch D38978 to the LLVM NVPTX backend which ensures the lowering of shared variables to an device memory which allows the sharing of references; - Patch (coming soon) is a patch to libomptarget runtime library which ensures that a list of references to shared variables is properly maintained. A simple code snippet which illustrates an implicit data sharing situation is as follows: ``` #pragma omp target { // master thread only int v; #pragma omp parallel { // worker threads // use v } } ``` Variable v is implicitly shared from the team master thread which executes the code in between the target and parallel directives. The worker threads must operate on the latest version of v, including any updates performed by the master. The code generated in this patch relies on the LLVM NVPTX patch (mentioned above) which prevents v from being lowered in the thread local memory of the master thread thus making the reference to this variable un-shareable with the workers. This ensures that the code generated by this patch is correct. Since the parallel region is outlined the passing of arguments to the outlined regions must preserve the original order of arguments. The runtime therefore maintains a list of references to shared variables thus ensuring their passing in the correct order. The passing of arguments to the outlined parallel function is performed in a separate function which the data sharing infrastructure constructs in this patch. The function is inlined when optimizations are enabled. Reviewers: hfinkel, carlo.bertolli, arpith-jacob, Hahnfeld, ABataev, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D38976 llvm-svn: 318773
2017-11-21 23:54:54 +08:00
// CK1: [[SHAREDARGS:%.+]] = alloca i8**
[OpenMP] Further adjustments of nvptx runtime functions Pass in default value of 1, similar to previous commit r318836. Differential Revision: https://reviews.llvm.org/D41012 llvm-svn: 321486
2017-12-27 18:39:56 +08:00
// CK1: call i1 @__kmpc_kernel_parallel(i8** %work_fn, i8*** [[SHAREDARGS]], i16 1)
[OpenMP] Add implicit data sharing support when offloading to NVIDIA GPUs using OpenMP device offloading Summary: This patch is part of the development effort to add support in the current OpenMP GPU offloading implementation for implicitly sharing variables between a target region executed by the team master thread and the worker threads within that team. This patch is the first of three required for successfully performing the implicit sharing of master thread variables with the worker threads within a team. The remaining two patches are: - Patch D38978 to the LLVM NVPTX backend which ensures the lowering of shared variables to an device memory which allows the sharing of references; - Patch (coming soon) is a patch to libomptarget runtime library which ensures that a list of references to shared variables is properly maintained. A simple code snippet which illustrates an implicit data sharing situation is as follows: ``` #pragma omp target { // master thread only int v; #pragma omp parallel { // worker threads // use v } } ``` Variable v is implicitly shared from the team master thread which executes the code in between the target and parallel directives. The worker threads must operate on the latest version of v, including any updates performed by the master. The code generated in this patch relies on the LLVM NVPTX patch (mentioned above) which prevents v from being lowered in the thread local memory of the master thread thus making the reference to this variable un-shareable with the workers. This ensures that the code generated by this patch is correct. Since the parallel region is outlined the passing of arguments to the outlined regions must preserve the original order of arguments. The runtime therefore maintains a list of references to shared variables thus ensuring their passing in the correct order. The passing of arguments to the outlined parallel function is performed in a separate function which the data sharing infrastructure constructs in this patch. The function is inlined when optimizations are enabled. Reviewers: hfinkel, carlo.bertolli, arpith-jacob, Hahnfeld, ABataev, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D38976 llvm-svn: 318773
2017-11-21 23:54:54 +08:00
// CK1: [[SHARGSTMP:%.+]] = load i8**, i8*** [[SHAREDARGS]]
Fix test/OpenMP/nvptx_data_sharing.cpp This was an oversight that stayed in the test from development. llvm-svn: 318779
2017-11-22 00:49:11 +08:00
// CK1: call void @__omp_outlined___wrapper{{.*}}({{.*}}, i8** [[SHARGSTMP]])
[OpenMP] Add implicit data sharing support when offloading to NVIDIA GPUs using OpenMP device offloading Summary: This patch is part of the development effort to add support in the current OpenMP GPU offloading implementation for implicitly sharing variables between a target region executed by the team master thread and the worker threads within that team. This patch is the first of three required for successfully performing the implicit sharing of master thread variables with the worker threads within a team. The remaining two patches are: - Patch D38978 to the LLVM NVPTX backend which ensures the lowering of shared variables to an device memory which allows the sharing of references; - Patch (coming soon) is a patch to libomptarget runtime library which ensures that a list of references to shared variables is properly maintained. A simple code snippet which illustrates an implicit data sharing situation is as follows: ``` #pragma omp target { // master thread only int v; #pragma omp parallel { // worker threads // use v } } ``` Variable v is implicitly shared from the team master thread which executes the code in between the target and parallel directives. The worker threads must operate on the latest version of v, including any updates performed by the master. The code generated in this patch relies on the LLVM NVPTX patch (mentioned above) which prevents v from being lowered in the thread local memory of the master thread thus making the reference to this variable un-shareable with the workers. This ensures that the code generated by this patch is correct. Since the parallel region is outlined the passing of arguments to the outlined regions must preserve the original order of arguments. The runtime therefore maintains a list of references to shared variables thus ensuring their passing in the correct order. The passing of arguments to the outlined parallel function is performed in a separate function which the data sharing infrastructure constructs in this patch. The function is inlined when optimizations are enabled. Reviewers: hfinkel, carlo.bertolli, arpith-jacob, Hahnfeld, ABataev, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D38976 llvm-svn: 318773
2017-11-21 23:54:54 +08:00
/// ========= In the kernel function ========= ///
[OpenMP] Add function attribute for triggering data sharing. Summary: The backend should only emit data sharing code for the cases where it is needed. A new function attribute is used by Clang to enable data sharing only for the cases where OpenMP semantics require it and there are variables that need to be shared. Reviewers: hfinkel, Hahnfeld, ABataev, carlo.bertolli, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D41123 llvm-svn: 320527
2017-12-13 05:38:43 +08:00
// CK1: {{.*}}define void @__omp_offloading{{.*}}test_ds{{.*}}() [[ATTR2:#.*]] {
[OpenMP] Add implicit data sharing support when offloading to NVIDIA GPUs using OpenMP device offloading Summary: This patch is part of the development effort to add support in the current OpenMP GPU offloading implementation for implicitly sharing variables between a target region executed by the team master thread and the worker threads within that team. This patch is the first of three required for successfully performing the implicit sharing of master thread variables with the worker threads within a team. The remaining two patches are: - Patch D38978 to the LLVM NVPTX backend which ensures the lowering of shared variables to an device memory which allows the sharing of references; - Patch (coming soon) is a patch to libomptarget runtime library which ensures that a list of references to shared variables is properly maintained. A simple code snippet which illustrates an implicit data sharing situation is as follows: ``` #pragma omp target { // master thread only int v; #pragma omp parallel { // worker threads // use v } } ``` Variable v is implicitly shared from the team master thread which executes the code in between the target and parallel directives. The worker threads must operate on the latest version of v, including any updates performed by the master. The code generated in this patch relies on the LLVM NVPTX patch (mentioned above) which prevents v from being lowered in the thread local memory of the master thread thus making the reference to this variable un-shareable with the workers. This ensures that the code generated by this patch is correct. Since the parallel region is outlined the passing of arguments to the outlined regions must preserve the original order of arguments. The runtime therefore maintains a list of references to shared variables thus ensuring their passing in the correct order. The passing of arguments to the outlined parallel function is performed in a separate function which the data sharing infrastructure constructs in this patch. The function is inlined when optimizations are enabled. Reviewers: hfinkel, carlo.bertolli, arpith-jacob, Hahnfeld, ABataev, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D38976 llvm-svn: 318773
2017-11-21 23:54:54 +08:00
// CK1: [[SHAREDARGS1:%.+]] = alloca i8**
[OpenMP] Further adjustments of nvptx runtime functions Pass in default value of 1, similar to previous commit r318836. Differential Revision: https://reviews.llvm.org/D41012 llvm-svn: 321486
2017-12-27 18:39:56 +08:00
// CK1: call void @__kmpc_kernel_prepare_parallel({{.*}}, i8*** [[SHAREDARGS1]], i32 1, i16 1)
[OpenMP] Add implicit data sharing support when offloading to NVIDIA GPUs using OpenMP device offloading Summary: This patch is part of the development effort to add support in the current OpenMP GPU offloading implementation for implicitly sharing variables between a target region executed by the team master thread and the worker threads within that team. This patch is the first of three required for successfully performing the implicit sharing of master thread variables with the worker threads within a team. The remaining two patches are: - Patch D38978 to the LLVM NVPTX backend which ensures the lowering of shared variables to an device memory which allows the sharing of references; - Patch (coming soon) is a patch to libomptarget runtime library which ensures that a list of references to shared variables is properly maintained. A simple code snippet which illustrates an implicit data sharing situation is as follows: ``` #pragma omp target { // master thread only int v; #pragma omp parallel { // worker threads // use v } } ``` Variable v is implicitly shared from the team master thread which executes the code in between the target and parallel directives. The worker threads must operate on the latest version of v, including any updates performed by the master. The code generated in this patch relies on the LLVM NVPTX patch (mentioned above) which prevents v from being lowered in the thread local memory of the master thread thus making the reference to this variable un-shareable with the workers. This ensures that the code generated by this patch is correct. Since the parallel region is outlined the passing of arguments to the outlined regions must preserve the original order of arguments. The runtime therefore maintains a list of references to shared variables thus ensuring their passing in the correct order. The passing of arguments to the outlined parallel function is performed in a separate function which the data sharing infrastructure constructs in this patch. The function is inlined when optimizations are enabled. Reviewers: hfinkel, carlo.bertolli, arpith-jacob, Hahnfeld, ABataev, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D38976 llvm-svn: 318773
2017-11-21 23:54:54 +08:00
// CK1: [[SHARGSTMP1:%.+]] = load i8**, i8*** [[SHAREDARGS1]]
// CK1: [[SHARGSTMP2:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP1]]
// CK1: [[SHAREDVAR:%.+]] = bitcast i32* {{.*}} to i8*
// CK1: store i8* [[SHAREDVAR]], i8** [[SHARGSTMP2]]
/// ========= In the data sharing wrapper function ========= ///
[OpenMP] Add function attribute for triggering data sharing. Summary: The backend should only emit data sharing code for the cases where it is needed. A new function attribute is used by Clang to enable data sharing only for the cases where OpenMP semantics require it and there are variables that need to be shared. Reviewers: hfinkel, Hahnfeld, ABataev, carlo.bertolli, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D41123 llvm-svn: 320527
2017-12-13 05:38:43 +08:00
// CK1: {{.*}}define internal void @__omp_outlined___wrapper({{.*}}i8**) [[ATTR1]] {
[OpenMP] Add implicit data sharing support when offloading to NVIDIA GPUs using OpenMP device offloading Summary: This patch is part of the development effort to add support in the current OpenMP GPU offloading implementation for implicitly sharing variables between a target region executed by the team master thread and the worker threads within that team. This patch is the first of three required for successfully performing the implicit sharing of master thread variables with the worker threads within a team. The remaining two patches are: - Patch D38978 to the LLVM NVPTX backend which ensures the lowering of shared variables to an device memory which allows the sharing of references; - Patch (coming soon) is a patch to libomptarget runtime library which ensures that a list of references to shared variables is properly maintained. A simple code snippet which illustrates an implicit data sharing situation is as follows: ``` #pragma omp target { // master thread only int v; #pragma omp parallel { // worker threads // use v } } ``` Variable v is implicitly shared from the team master thread which executes the code in between the target and parallel directives. The worker threads must operate on the latest version of v, including any updates performed by the master. The code generated in this patch relies on the LLVM NVPTX patch (mentioned above) which prevents v from being lowered in the thread local memory of the master thread thus making the reference to this variable un-shareable with the workers. This ensures that the code generated by this patch is correct. Since the parallel region is outlined the passing of arguments to the outlined regions must preserve the original order of arguments. The runtime therefore maintains a list of references to shared variables thus ensuring their passing in the correct order. The passing of arguments to the outlined parallel function is performed in a separate function which the data sharing infrastructure constructs in this patch. The function is inlined when optimizations are enabled. Reviewers: hfinkel, carlo.bertolli, arpith-jacob, Hahnfeld, ABataev, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D38976 llvm-svn: 318773
2017-11-21 23:54:54 +08:00
// CK1: [[SHAREDARGS2:%.+]] = alloca i8**
// CK1: store i8** %2, i8*** [[SHAREDARGS2]]
// CK1: [[SHARGSTMP3:%.+]] = load i8**, i8*** [[SHAREDARGS2]]
// CK1: [[SHARGSTMP4:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP3]]
// CK1: [[SHARGSTMP5:%.+]] = bitcast i8** [[SHARGSTMP4]] to i32**
// CK1: [[SHARGSTMP6:%.+]] = load i32*, i32** [[SHARGSTMP5]]
// CK1: call void @__omp_outlined__({{.*}}, i32* [[SHARGSTMP6]])
[OpenMP] Add function attribute for triggering data sharing. Summary: The backend should only emit data sharing code for the cases where it is needed. A new function attribute is used by Clang to enable data sharing only for the cases where OpenMP semantics require it and there are variables that need to be shared. Reviewers: hfinkel, Hahnfeld, ABataev, carlo.bertolli, caomhin Reviewed By: ABataev Subscribers: cfe-commits, jholewinski Differential Revision: https://reviews.llvm.org/D41123 llvm-svn: 320527
2017-12-13 05:38:43 +08:00
/// ========= Attributes ========= ///
// CK1-NOT: attributes [[ATTR1]] = { {{.*}}"has-nvptx-shared-depot"{{.*}} }
// CK1: attributes [[ATTR2]] = { {{.*}}"has-nvptx-shared-depot"{{.*}} }
Fix test/OpenMP/nvptx_data_sharing.cpp This was an oversight that stayed in the test from development. llvm-svn: 318779
2017-11-22 00:49:11 +08:00
#endif