RE: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes
Hi, What I mean is the spec of spir 1.2/2.0 has very clear example about how the metadata should organized. But now with this patch, the result of spir is changed and not like it before. So what I want to know is why we change the old spir example style code, but not add something alongside. Now we have metadata after each function but spir shows a opencl.kernels metadata is need for all modules: "Each SPIR module has a opencl.kernels named metadata node containing a list of metadata objects." I think this patch maybe useful for some other target, but we should not break the old spir as now we have not had a substitute one. Thanks Xiuli -Original Message- From: Anastasia Stulova [mailto:anastasia.stul...@arm.com] Sent: Tuesday, August 2, 2016 2:14 AM To: Pan Xiuli <xiuli...@outlook.com>; reviews+d20979+public+e2872c9c869f1...@reviews.llvm.org; alexey.ba...@intel.com; Liu, Yaxun (Sam) (yaxun@amd.com) <yaxun@amd.com> Cc: thomas.stell...@amd.com; cfe-commits@lists.llvm.org; nd <n...@arm.com> Subject: RE: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes Hi Xiuli, Could you please elaborate what do you think it broken exactly? Because we haven't actually removed opencl.kernels metadata but just changed the format of it. Basically, we are using function metadata instead of generic metadata as it was before. Thanks, Anastasia -Original Message- From: Pan Xiuli [mailto:xiuli...@outlook.com] Sent: 01 August 2016 05:54 To: reviews+d20979+public+e2872c9c869f1...@reviews.llvm.org; Anastasia Stulova; alexey.ba...@intel.com Cc: thomas.stell...@amd.com; cfe-commits@lists.llvm.org Subject: RE: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes Hi Sam, I am now using llvm39 to enable our opencl driver, but I found this patch broke the spir target as well. The opencl.kernels metadata is required by spir spec. I think we should keep the old code for spir target. Thanks Xiuli -Original Message- From: Yaxun Liu [mailto:yaxun@amd.com] Sent: Wednesday, June 22, 2016 11:04 PM To: yaxun@amd.com; xiuli...@outlook.com; anastasia.stul...@arm.com; alexey.ba...@intel.com Cc: thomas.stell...@amd.com; cfe-commits@lists.llvm.org Subject: Re: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes This revision was automatically updated to reflect the committed changes. Closed by commit rL273425: [OpenCL] Use function metadata to represent kernel attributes (authored by yaxunl). Changed prior to commit: http://reviews.llvm.org/D20979?vs=60545=61555#toc Repository: rL LLVM http://reviews.llvm.org/D20979 Files: cfe/trunk/lib/CodeGen/CodeGenFunction.cpp cfe/trunk/test/CodeGenOpenCL/kernel-arg-info.cl cfe/trunk/test/CodeGenOpenCL/kernel-attributes.cl cfe/trunk/test/CodeGenOpenCL/kernel-metadata.cl ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
RE: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes
Hi Xiuli, Could you please elaborate what do you think it broken exactly? Because we haven't actually removed opencl.kernels metadata but just changed the format of it. Basically, we are using function metadata instead of generic metadata as it was before. Thanks, Anastasia -Original Message- From: Pan Xiuli [mailto:xiuli...@outlook.com] Sent: 01 August 2016 05:54 To: reviews+d20979+public+e2872c9c869f1...@reviews.llvm.org; Anastasia Stulova; alexey.ba...@intel.com Cc: thomas.stell...@amd.com; cfe-commits@lists.llvm.org Subject: RE: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes Hi Sam, I am now using llvm39 to enable our opencl driver, but I found this patch broke the spir target as well. The opencl.kernels metadata is required by spir spec. I think we should keep the old code for spir target. Thanks Xiuli -Original Message- From: Yaxun Liu [mailto:yaxun@amd.com] Sent: Wednesday, June 22, 2016 11:04 PM To: yaxun@amd.com; xiuli...@outlook.com; anastasia.stul...@arm.com; alexey.ba...@intel.com Cc: thomas.stell...@amd.com; cfe-commits@lists.llvm.org Subject: Re: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes This revision was automatically updated to reflect the committed changes. Closed by commit rL273425: [OpenCL] Use function metadata to represent kernel attributes (authored by yaxunl). Changed prior to commit: http://reviews.llvm.org/D20979?vs=60545=61555#toc Repository: rL LLVM http://reviews.llvm.org/D20979 Files: cfe/trunk/lib/CodeGen/CodeGenFunction.cpp cfe/trunk/test/CodeGenOpenCL/kernel-arg-info.cl cfe/trunk/test/CodeGenOpenCL/kernel-attributes.cl cfe/trunk/test/CodeGenOpenCL/kernel-metadata.cl ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
RE: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes
Hi Sam, I am now using llvm39 to enable our opencl driver, but I found this patch broke the spir target as well. The opencl.kernels metadata is required by spir spec. I think we should keep the old code for spir target. Thanks Xiuli -Original Message- From: Yaxun Liu [mailto:yaxun@amd.com] Sent: Wednesday, June 22, 2016 11:04 PM To: yaxun@amd.com; xiuli...@outlook.com; anastasia.stul...@arm.com; alexey.ba...@intel.com Cc: thomas.stell...@amd.com; cfe-commits@lists.llvm.org Subject: Re: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes This revision was automatically updated to reflect the committed changes. Closed by commit rL273425: [OpenCL] Use function metadata to represent kernel attributes (authored by yaxunl). Changed prior to commit: http://reviews.llvm.org/D20979?vs=60545=61555#toc Repository: rL LLVM http://reviews.llvm.org/D20979 Files: cfe/trunk/lib/CodeGen/CodeGenFunction.cpp cfe/trunk/test/CodeGenOpenCL/kernel-arg-info.cl cfe/trunk/test/CodeGenOpenCL/kernel-attributes.cl cfe/trunk/test/CodeGenOpenCL/kernel-metadata.cl ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
Re: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes
This revision was automatically updated to reflect the committed changes.
Closed by commit rL273425: [OpenCL] Use function metadata to represent kernel
attributes (authored by yaxunl).
Changed prior to commit:
http://reviews.llvm.org/D20979?vs=60545=61555#toc
Repository:
rL LLVM
http://reviews.llvm.org/D20979
Files:
cfe/trunk/lib/CodeGen/CodeGenFunction.cpp
cfe/trunk/test/CodeGenOpenCL/kernel-arg-info.cl
cfe/trunk/test/CodeGenOpenCL/kernel-attributes.cl
cfe/trunk/test/CodeGenOpenCL/kernel-metadata.cl
Index: cfe/trunk/test/CodeGenOpenCL/kernel-attributes.cl
===
--- cfe/trunk/test/CodeGenOpenCL/kernel-attributes.cl
+++ cfe/trunk/test/CodeGenOpenCL/kernel-attributes.cl
@@ -3,14 +3,12 @@
typedef unsigned int uint4 __attribute__((ext_vector_type(4)));
kernel __attribute__((vec_type_hint(int))) __attribute__((reqd_work_group_size(1,2,4))) void kernel1(int a) {}
+// CHECK: define void @kernel1(i32 %a) {{[^{]+}} !vec_type_hint ![[MD1:[0-9]+]] !reqd_work_group_size ![[MD2:[0-9]+]]
kernel __attribute__((vec_type_hint(uint4))) __attribute__((work_group_size_hint(8,16,32))) void kernel2(int a) {}
+// CHECK: define void @kernel2(i32 %a) {{[^{]+}} !vec_type_hint ![[MD3:[0-9]+]] !work_group_size_hint ![[MD4:[0-9]+]]
-// CHECK: opencl.kernels = !{[[MDNODE0:![0-9]+]], [[MDNODE3:![0-9]+]]}
-
-// CHECK: [[MDNODE0]] = !{void (i32)* @kernel1, {{.*}} [[MDNODE1:![0-9]+]], [[MDNODE2:![0-9]+]]}
-// CHECK: [[MDNODE1]] = !{!"vec_type_hint", i32 undef, i32 1}
-// CHECK: [[MDNODE2]] = !{!"reqd_work_group_size", i32 1, i32 2, i32 4}
-// CHECK: [[MDNODE3]] = !{void (i32)* @kernel2, {{.*}} [[MDNODE4:![0-9]+]], [[MDNODE5:![0-9]+]]}
-// CHECK: [[MDNODE4]] = !{!"vec_type_hint", <4 x i32> undef, i32 0}
-// CHECK: [[MDNODE5]] = !{!"work_group_size_hint", i32 8, i32 16, i32 32}
+// CHECK: [[MD1]] = !{i32 undef, i32 1}
+// CHECK: [[MD2]] = !{i32 1, i32 2, i32 4}
+// CHECK: [[MD3]] = !{<4 x i32> undef, i32 0}
+// CHECK: [[MD4]] = !{i32 8, i32 16, i32 32}
Index: cfe/trunk/test/CodeGenOpenCL/kernel-arg-info.cl
===
--- cfe/trunk/test/CodeGenOpenCL/kernel-arg-info.cl
+++ cfe/trunk/test/CodeGenOpenCL/kernel-arg-info.cl
@@ -1,55 +1,88 @@
-// RUN: %clang_cc1 %s -cl-kernel-arg-info -emit-llvm -o - -triple spir-unknown-unknown | FileCheck %s -check-prefix ARGINFO
-// RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown | FileCheck %s -check-prefix NO-ARGINFO
+// RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown | FileCheck %s
+// RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown -cl-kernel-arg-info | FileCheck %s -check-prefix ARGINFO
kernel void foo(__global int * restrict X, const int Y,
volatile int anotherArg, __constant float * restrict Z) {
*X = Y + anotherArg;
}
-
-// CHECK: !{!"kernel_arg_addr_space", i32 1, i32 0, i32 0, i32 2}
-// CHECK: !{!"kernel_arg_access_qual", !"none", !"none", !"none", !"none"}
-// CHECK: !{!"kernel_arg_type", !"int*", !"int", !"int", !"float*"}
-// CHECK: !{!"kernel_arg_base_type", !"int*", !"int", !"int", !"float*"}
-// CHECK: !{!"kernel_arg_type_qual", !"restrict", !"const", !"volatile", !"restrict const"}
-// ARGINFO: !{!"kernel_arg_name", !"X", !"Y", !"anotherArg", !"Z"}
-// NO-ARGINFO-NOT: !{!"kernel_arg_name", !"X", !"Y", !"anotherArg", !"Z"}
+// CHECK: define spir_kernel void @foo{{[^!]+}}
+// CHECK: !kernel_arg_addr_space ![[MD11:[0-9]+]]
+// CHECK: !kernel_arg_access_qual ![[MD12:[0-9]+]]
+// CHECK: !kernel_arg_type ![[MD13:[0-9]+]]
+// CHECK: !kernel_arg_base_type ![[MD13]]
+// CHECK: !kernel_arg_type_qual ![[MD14:[0-9]+]]
+// CHECK-NOT: !kernel_arg_name
+// ARGINFO: !kernel_arg_name ![[MD15:[0-9]+]]
kernel void foo2(read_only image1d_t img1, image2d_t img2, write_only image2d_array_t img3) {
}
-// CHECK: !{!"kernel_arg_addr_space", i32 1, i32 1, i32 1}
-// CHECK: !{!"kernel_arg_access_qual", !"read_only", !"read_only", !"write_only"}
-// CHECK: !{!"kernel_arg_type", !"image1d_t", !"image2d_t", !"image2d_array_t"}
-// CHECK: !{!"kernel_arg_base_type", !"image1d_t", !"image2d_t", !"image2d_array_t"}
-// CHECK: !{!"kernel_arg_type_qual", !"", !"", !""}
-// ARGINFO: !{!"kernel_arg_name", !"img1", !"img2", !"img3"}
-// NO-ARGINFO-NOT: !{!"kernel_arg_name", !"img1", !"img2", !"img3"}
+// CHECK: define spir_kernel void @foo2{{[^!]+}}
+// CHECK: !kernel_arg_addr_space ![[MD21:[0-9]+]]
+// CHECK: !kernel_arg_access_qual ![[MD22:[0-9]+]]
+// CHECK: !kernel_arg_type ![[MD23:[0-9]+]]
+// CHECK: !kernel_arg_base_type ![[MD23]]
+// CHECK: !kernel_arg_type_qual ![[MD24:[0-9]+]]
+// CHECK-NOT: !kernel_arg_name
+// ARGINFO: !kernel_arg_name ![[MD25:[0-9]+]]
kernel void foo3(__global half * X) {
}
-// CHECK: !{!"kernel_arg_addr_space", i32 1}
-// CHECK: !{!"kernel_arg_access_qual", !"none"}
-// CHECK: !{!"kernel_arg_type", !"half*"}
-// CHECK: !{!"kernel_arg_base_type", !"half*"}
-//
Re: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes
yaxunl added inline comments.
Comment at: test/CodeGenOpenCL/kernel-attributes.cl:9
@@ -8,3 +10,1 @@
-// CHECK: opencl.kernels = !{[[MDNODE0:![0-9]+]], [[MDNODE3:![0-9]+]]}
-
bader wrote:
> AFAIK, this named metadata node were also useful for kernel functions
> traversal.
> What is the new BKM to differentiate kernel/non-kernel functions?
> Check for !kernel_arg_* function attribute?
I think it could be used for that purpose, since kernel functions and only
kernel functions have this function attribute.
http://reviews.llvm.org/D20979
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
Re: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes
bader accepted this revision.
bader added a comment.
LGTM.
Comment at: test/CodeGenOpenCL/kernel-attributes.cl:9
@@ -8,3 +10,1 @@
-// CHECK: opencl.kernels = !{[[MDNODE0:![0-9]+]], [[MDNODE3:![0-9]+]]}
-
AFAIK, this named metadata node were also useful for kernel functions traversal.
What is the new BKM to differentiate kernel/non-kernel functions?
Check for !kernel_arg_* function attribute?
http://reviews.llvm.org/D20979
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
Re: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes
yaxunl added a comment. Ping! Alexey/Xiuli, are you OK with this change? Thanks. http://reviews.llvm.org/D20979 ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
Re: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes
Anastasia accepted this revision. Anastasia added a comment. This revision is now accepted and ready to land. LGTM! http://reviews.llvm.org/D20979 ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
Re: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes
yaxunl added a comment. I updated the patch to use function metadata only. Please review. Thanks. http://reviews.llvm.org/D20979 ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
Re: [PATCH] D20979: [OpenCL] Use function metadata to represent kernel attributes
yaxunl retitled this revision from "[OpenCL] Use function attribute/metadata to
represent kernel attributes" to "[OpenCL] Use function metadata to represent
kernel attributes".
yaxunl updated the summary for this revision.
yaxunl updated this revision to Diff 60545.
http://reviews.llvm.org/D20979
Files:
lib/CodeGen/CodeGenFunction.cpp
test/CodeGenOpenCL/kernel-arg-info.cl
test/CodeGenOpenCL/kernel-attributes.cl
test/CodeGenOpenCL/kernel-metadata.cl
Index: test/CodeGenOpenCL/kernel-metadata.cl
===
--- test/CodeGenOpenCL/kernel-metadata.cl
+++ test/CodeGenOpenCL/kernel-metadata.cl
@@ -6,10 +6,5 @@
__kernel void kernel_function() {
}
-// CHECK: !opencl.kernels = !{!0}
-// CHECK: !0 = !{void ()* @kernel_function, !1, !2, !3, !4, !5}
-// CHECK: !1 = !{!"kernel_arg_addr_space"}
-// CHECK: !2 = !{!"kernel_arg_access_qual"}
-// CHECK: !3 = !{!"kernel_arg_type"}
-// CHECK: !4 = !{!"kernel_arg_base_type"}
-// CHECK: !5 = !{!"kernel_arg_type_qual"}
+// CHECK: define void @kernel_function() #{{[0-9]+}} !kernel_arg_addr_space ![[MD:[0-9]+]] !kernel_arg_access_qual ![[MD]] !kernel_arg_type ![[MD]] !kernel_arg_base_type ![[MD]] !kernel_arg_type_qual ![[MD]] {
+// CHECK: ![[MD]] = !{}
Index: test/CodeGenOpenCL/kernel-attributes.cl
===
--- test/CodeGenOpenCL/kernel-attributes.cl
+++ test/CodeGenOpenCL/kernel-attributes.cl
@@ -3,14 +3,12 @@
typedef unsigned int uint4 __attribute__((ext_vector_type(4)));
kernel __attribute__((vec_type_hint(int))) __attribute__((reqd_work_group_size(1,2,4))) void kernel1(int a) {}
+// CHECK: define void @kernel1(i32 %a) #{{[^{]+}} !vec_type_hint ![[MD1:[0-9]+]] !reqd_work_group_size ![[MD2:[0-9]+]]
kernel __attribute__((vec_type_hint(uint4))) __attribute__((work_group_size_hint(8,16,32))) void kernel2(int a) {}
+// CHECK: define void @kernel2(i32 %a) #{{[^{]+}} !vec_type_hint ![[MD3:[0-9]+]] !work_group_size_hint ![[MD4:[0-9]+]]
-// CHECK: opencl.kernels = !{[[MDNODE0:![0-9]+]], [[MDNODE3:![0-9]+]]}
-
-// CHECK: [[MDNODE0]] = !{void (i32)* @kernel1, {{.*}} [[MDNODE1:![0-9]+]], [[MDNODE2:![0-9]+]]}
-// CHECK: [[MDNODE1]] = !{!"vec_type_hint", i32 undef, i32 1}
-// CHECK: [[MDNODE2]] = !{!"reqd_work_group_size", i32 1, i32 2, i32 4}
-// CHECK: [[MDNODE3]] = !{void (i32)* @kernel2, {{.*}} [[MDNODE4:![0-9]+]], [[MDNODE5:![0-9]+]]}
-// CHECK: [[MDNODE4]] = !{!"vec_type_hint", <4 x i32> undef, i32 0}
-// CHECK: [[MDNODE5]] = !{!"work_group_size_hint", i32 8, i32 16, i32 32}
+// CHECK: [[MD1]] = !{i32 undef, i32 1}
+// CHECK: [[MD2]] = !{i32 1, i32 2, i32 4}
+// CHECK: [[MD3]] = !{<4 x i32> undef, i32 0}
+// CHECK: [[MD4]] = !{i32 8, i32 16, i32 32}
Index: test/CodeGenOpenCL/kernel-arg-info.cl
===
--- test/CodeGenOpenCL/kernel-arg-info.cl
+++ test/CodeGenOpenCL/kernel-arg-info.cl
@@ -1,55 +1,88 @@
-// RUN: %clang_cc1 %s -cl-kernel-arg-info -emit-llvm -o - -triple spir-unknown-unknown | FileCheck %s -check-prefix ARGINFO
-// RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown | FileCheck %s -check-prefix NO-ARGINFO
+// RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown | FileCheck %s
+// RUN: %clang_cc1 %s -emit-llvm -o - -triple spir-unknown-unknown -cl-kernel-arg-info | FileCheck %s -check-prefix ARGINFO
kernel void foo(__global int * restrict X, const int Y,
volatile int anotherArg, __constant float * restrict Z) {
*X = Y + anotherArg;
}
-
-// CHECK: !{!"kernel_arg_addr_space", i32 1, i32 0, i32 0, i32 2}
-// CHECK: !{!"kernel_arg_access_qual", !"none", !"none", !"none", !"none"}
-// CHECK: !{!"kernel_arg_type", !"int*", !"int", !"int", !"float*"}
-// CHECK: !{!"kernel_arg_base_type", !"int*", !"int", !"int", !"float*"}
-// CHECK: !{!"kernel_arg_type_qual", !"restrict", !"const", !"volatile", !"restrict const"}
-// ARGINFO: !{!"kernel_arg_name", !"X", !"Y", !"anotherArg", !"Z"}
-// NO-ARGINFO-NOT: !{!"kernel_arg_name", !"X", !"Y", !"anotherArg", !"Z"}
+// CHECK: define spir_kernel void @foo{{[^!]+}}
+// CHECK: !kernel_arg_addr_space ![[MD11:[0-9]+]]
+// CHECK: !kernel_arg_access_qual ![[MD12:[0-9]+]]
+// CHECK: !kernel_arg_type ![[MD13:[0-9]+]]
+// CHECK: !kernel_arg_base_type ![[MD13]]
+// CHECK: !kernel_arg_type_qual ![[MD14:[0-9]+]]
+// CHECK-NOT: !kernel_arg_name
+// ARGINFO: !kernel_arg_name ![[MD15:[0-9]+]]
kernel void foo2(read_only image1d_t img1, image2d_t img2, write_only image2d_array_t img3) {
}
-// CHECK: !{!"kernel_arg_addr_space", i32 1, i32 1, i32 1}
-// CHECK: !{!"kernel_arg_access_qual", !"read_only", !"read_only", !"write_only"}
-// CHECK: !{!"kernel_arg_type", !"image1d_t", !"image2d_t", !"image2d_array_t"}
-// CHECK: !{!"kernel_arg_base_type", !"image1d_t", !"image2d_t", !"image2d_array_t"}
-// CHECK: !{!"kernel_arg_type_qual", !"", !"", !""}
-// ARGINFO: !{!"kernel_arg_name", !"img1", !"img2",
