[Libclc-dev] [PATCH 1/5] Restore support for llvm-3.9
Aaron Watry via Libclc-dev
libclc-dev at lists.llvm.org
Thu Sep 28 18:42:40 PDT 2017
For patches 1, 2, 5, and 6: Acked-by: Aaron Watry <awatry at gmail.com>
Patch 3 is Reviewed-by: Aaron Watry <awatry at gmail.com>
I've got one small comment on patch 4 that I'll send in a moment.
--Aaron
On Wed, Sep 20, 2017 at 11:06 PM, Jan Vesely via Libclc-dev
<libclc-dev at lists.llvm.org> wrote:
> Signed-off-by: Jan Vesely <jan.vesely at rutgers.edu>
> ---
> I haven't bothered fixing amdgcn-amdhsa beyond successful build,
> it produces invalid calls to workitem postiion similar to bug #99856.
>
> Please do consider this. there are still plenty of people running
> llvm-3.9.
>
> Jan
>
> amdgcn-amdhsa/lib/OVERRIDES | 3 +++
> amdgcn/lib/OVERRIDES_3.9 | 3 +++
> amdgcn/lib/SOURCES_3.9 | 3 +++
> amdgcn/lib/workitem/get_global_size.39.ll | 18 ++++++++++++++++++
> amdgcn/lib/workitem/get_local_size.39.ll | 18 ++++++++++++++++++
> amdgcn/lib/workitem/get_num_groups.39.ll | 18 ++++++++++++++++++
> amdgpu/lib/SOURCES_3.9 | 2 ++
> configure.py | 22 +++++++++++++++++-----
> utils/prepare-builtins.cpp | 9 +++++++++
> 9 files changed, 91 insertions(+), 5 deletions(-)
> create mode 100644 amdgcn/lib/OVERRIDES_3.9
> create mode 100644 amdgcn/lib/SOURCES_3.9
> create mode 100644 amdgcn/lib/workitem/get_global_size.39.ll
> create mode 100644 amdgcn/lib/workitem/get_local_size.39.ll
> create mode 100644 amdgcn/lib/workitem/get_num_groups.39.ll
> create mode 100644 amdgpu/lib/SOURCES_3.9
>
> diff --git a/amdgcn-amdhsa/lib/OVERRIDES b/amdgcn-amdhsa/lib/OVERRIDES
> index c9bd69b..e4dfca1 100644
> --- a/amdgcn-amdhsa/lib/OVERRIDES
> +++ b/amdgcn-amdhsa/lib/OVERRIDES
> @@ -1 +1,4 @@
> workitem/get_num_groups.ll
> +workitem/get_global_size.39.ll
> +workitem/get_local_size.39.ll
> +workitem/get_num_groups.39.ll
> diff --git a/amdgcn/lib/OVERRIDES_3.9 b/amdgcn/lib/OVERRIDES_3.9
> new file mode 100644
> index 0000000..f26fbbe
> --- /dev/null
> +++ b/amdgcn/lib/OVERRIDES_3.9
> @@ -0,0 +1,3 @@
> +workitem/get_global_size.ll
> +workitem/get_local_size.ll
> +workitem/get_num_groups.ll
> diff --git a/amdgcn/lib/SOURCES_3.9 b/amdgcn/lib/SOURCES_3.9
> new file mode 100644
> index 0000000..6ee7cb8
> --- /dev/null
> +++ b/amdgcn/lib/SOURCES_3.9
> @@ -0,0 +1,3 @@
> +workitem/get_global_size.39.ll
> +workitem/get_local_size.39.ll
> +workitem/get_num_groups.39.ll
> diff --git a/amdgcn/lib/workitem/get_global_size.39.ll b/amdgcn/lib/workitem/get_global_size.39.ll
> new file mode 100644
> index 0000000..d51def3
> --- /dev/null
> +++ b/amdgcn/lib/workitem/get_global_size.39.ll
> @@ -0,0 +1,18 @@
> +declare i32 @llvm.r600.read.global.size.x() nounwind readnone
> +declare i32 @llvm.r600.read.global.size.y() nounwind readnone
> +declare i32 @llvm.r600.read.global.size.z() nounwind readnone
> +
> +define i32 @get_global_size(i32 %dim) nounwind readnone alwaysinline {
> + switch i32 %dim, label %default [i32 0, label %x_dim i32 1, label %y_dim i32 2, label %z_dim]
> +x_dim:
> + %x = call i32 @llvm.r600.read.global.size.x()
> + ret i32 %x
> +y_dim:
> + %y = call i32 @llvm.r600.read.global.size.y()
> + ret i32 %y
> +z_dim:
> + %z = call i32 @llvm.r600.read.global.size.z()
> + ret i32 %z
> +default:
> + ret i32 1
> +}
> diff --git a/amdgcn/lib/workitem/get_local_size.39.ll b/amdgcn/lib/workitem/get_local_size.39.ll
> new file mode 100644
> index 0000000..92c6d26
> --- /dev/null
> +++ b/amdgcn/lib/workitem/get_local_size.39.ll
> @@ -0,0 +1,18 @@
> +declare i32 @llvm.r600.read.local.size.x() nounwind readnone
> +declare i32 @llvm.r600.read.local.size.y() nounwind readnone
> +declare i32 @llvm.r600.read.local.size.z() nounwind readnone
> +
> +define i32 @get_local_size(i32 %dim) nounwind readnone alwaysinline {
> + switch i32 %dim, label %default [i32 0, label %x_dim i32 1, label %y_dim i32 2, label %z_dim]
> +x_dim:
> + %x = call i32 @llvm.r600.read.local.size.x()
> + ret i32 %x
> +y_dim:
> + %y = call i32 @llvm.r600.read.local.size.y()
> + ret i32 %y
> +z_dim:
> + %z = call i32 @llvm.r600.read.local.size.z()
> + ret i32 %z
> +default:
> + ret i32 1
> +}
> diff --git a/amdgcn/lib/workitem/get_num_groups.39.ll b/amdgcn/lib/workitem/get_num_groups.39.ll
> new file mode 100644
> index 0000000..fd5c17c
> --- /dev/null
> +++ b/amdgcn/lib/workitem/get_num_groups.39.ll
> @@ -0,0 +1,18 @@
> +declare i32 @llvm.r600.read.ngroups.x() nounwind readnone
> +declare i32 @llvm.r600.read.ngroups.y() nounwind readnone
> +declare i32 @llvm.r600.read.ngroups.z() nounwind readnone
> +
> +define i32 @get_num_groups(i32 %dim) nounwind readnone alwaysinline {
> + switch i32 %dim, label %default [i32 0, label %x_dim i32 1, label %y_dim i32 2, label %z_dim]
> +x_dim:
> + %x = call i32 @llvm.r600.read.ngroups.x()
> + ret i32 %x
> +y_dim:
> + %y = call i32 @llvm.r600.read.ngroups.y()
> + ret i32 %y
> +z_dim:
> + %z = call i32 @llvm.r600.read.ngroups.z()
> + ret i32 %z
> +default:
> + ret i32 1
> +}
> diff --git a/amdgpu/lib/SOURCES_3.9 b/amdgpu/lib/SOURCES_3.9
> new file mode 100644
> index 0000000..69c5e5c
> --- /dev/null
> +++ b/amdgpu/lib/SOURCES_3.9
> @@ -0,0 +1,2 @@
> +shared/vload_half_helpers.ll
> +shared/vstore_half_helpers.ll
> diff --git a/configure.py b/configure.py
> index cc4c84b..bcdc6d0 100755
> --- a/configure.py
> +++ b/configure.py
> @@ -72,8 +72,8 @@ llvm_version = llvm_config(['--version']).replace('svn', '').split('.')
> llvm_int_version = int(llvm_version[0]) * 100 + int(llvm_version[1]) * 10
> llvm_string_version = llvm_version[0] + '.' + llvm_version[1]
>
> -if llvm_int_version < 400:
> - print("libclc requires LLVM >= 4.0")
> +if llvm_int_version < 390:
> + print("libclc requires LLVM >= 3.9")
> sys.exit(1)
>
> llvm_system_libs = llvm_config(['--system-libs'])
> @@ -81,7 +81,7 @@ llvm_bindir = llvm_config(['--bindir'])
> llvm_core_libs = llvm_config(['--libs', 'core', 'bitreader', 'bitwriter']) + ' ' + \
> llvm_system_libs + ' ' + \
> llvm_config(['--ldflags'])
> -llvm_cxxflags = llvm_config(['--cxxflags']) + ' -fno-exceptions -fno-rtti'
> +llvm_cxxflags = llvm_config(['--cxxflags']) + ' -fno-exceptions -fno-rtti -DHAVE_LLVM=0x' + str(llvm_int_version)
> llvm_libdir = llvm_config(['--libdir'])
>
> llvm_clang = os.path.join(llvm_bindir, 'clang')
> @@ -108,9 +108,13 @@ available_targets = {
> 'nvptx64--nvidiacl' : { 'devices' : [{'gpu' : '', 'aliases' : []} ]},
> }
>
> -available_targets['amdgcn-mesa-mesa3d'] = available_targets['amdgcn--']
>
> -default_targets = ['nvptx--nvidiacl', 'nvptx64--nvidiacl', 'r600--', 'amdgcn--', 'amdgcn--amdhsa', 'amdgcn-mesa-mesa3d']
> +default_targets = ['nvptx--nvidiacl', 'nvptx64--nvidiacl', 'r600--', 'amdgcn--', 'amdgcn--amdhsa']
> +
> +#mesa is using amdgcn-mesa-mesa3d since llvm-4.0
> +if llvm_int_version > 390:
> + available_targets['amdgcn-mesa-mesa3d'] = available_targets['amdgcn--']
> + default_targets.append('amdgcn-mesa-mesa3d')
>
> targets = args
> if not targets:
> @@ -217,6 +221,8 @@ for target in targets:
> override_list_file = os.path.join(libdir, 'OVERRIDES')
> compat_list_file = os.path.join(libdir,
> 'SOURCES_' + llvm_string_version)
> + compat_list_override = os.path.join(libdir,
> + 'OVERRIDES_' + llvm_string_version)
>
> # Build compat list
> if os.path.exists(compat_list_file):
> @@ -224,6 +230,12 @@ for target in targets:
> compat = compat.rstrip()
> compats.append(compat)
>
> + # Add target compat overrides
> + if os.path.exists(compat_list_override):
> + for override in open(compat_list_override).readlines():
> + override = override.rstrip()
> + sources_seen.add(override)
> +
> # Add target overrides
> if os.path.exists(override_list_file):
> for override in open(override_list_file).readlines():
> diff --git a/utils/prepare-builtins.cpp b/utils/prepare-builtins.cpp
> index 882dadd..8022e85 100644
> --- a/utils/prepare-builtins.cpp
> +++ b/utils/prepare-builtins.cpp
> @@ -1,5 +1,10 @@
> +#if HAVE_LLVM > 0x0390
> #include "llvm/Bitcode/BitcodeReader.h"
> #include "llvm/Bitcode/BitcodeWriter.h"
> +#else
> +#include "llvm/Bitcode/ReaderWriter.h"
> +#endif
> +
> #include "llvm/IR/Function.h"
> #include "llvm/IR/GlobalVariable.h"
> #include "llvm/IR/LLVMContext.h"
> @@ -41,8 +46,12 @@ int main(int argc, char **argv) {
> } else {
> std::unique_ptr<MemoryBuffer> &BufferPtr = BufferOrErr.get();
> ErrorOr<std::unique_ptr<Module>> ModuleOrErr =
> +#if HAVE_LLVM > 0x0390
> expectedToErrorOrAndEmitErrors(Context,
> parseBitcodeFile(BufferPtr.get()->getMemBufferRef(), Context));
> +#else
> + parseBitcodeFile(BufferPtr.get()->getMemBufferRef(), Context);
> +#endif
> if (std::error_code ec = ModuleOrErr.getError())
> ErrorMessage = ec.message();
>
> --
> 2.13.5
>
> _______________________________________________
> Libclc-dev mailing list
> Libclc-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/libclc-dev
More information about the Libclc-dev
mailing list