sanitize qualified name for nvptx backend

This commit is contained in:
Guillaume Wenzek 2022-09-16 22:21:14 +02:00 committed by Andrew Kelley
parent 92a857b76c
commit aad983cf40
4 changed files with 26 additions and 23 deletions

View File

@ -951,6 +951,13 @@ pub const Target = struct {
};
}
pub fn isNvptx(arch: Arch) bool {
return switch (arch) {
.nvptx, .nvptx64 => true,
else => false,
};
}
pub fn parseCpuModel(arch: Arch, cpu_name: []const u8) !*const Cpu.Model {
for (arch.allCpuModels()) |cpu| {
if (mem.eql(u8, cpu_name, cpu.name)) {

View File

@ -720,6 +720,15 @@ pub const Decl = struct {
var buffer = std.ArrayList(u8).init(mod.gpa);
defer buffer.deinit();
try decl.renderFullyQualifiedName(mod, buffer.writer());
// Sanitize the name for nvptx which is more restrictive.
if (mod.comp.bin_file.options.target.cpu.arch.isNvptx()) {
for (buffer.items) |*byte| switch (byte.*) {
'{', '}', '*', '[', ']', '(', ')', ',', ' ', '\'' => byte.* = '_',
else => {},
};
}
return buffer.toOwnedSliceSentinel(0);
}

View File

@ -28,10 +28,7 @@ pub fn createEmpty(gpa: Allocator, options: link.Options) !*NvPtx {
if (!build_options.have_llvm) return error.PtxArchNotSupported;
if (!options.use_llvm) return error.PtxArchNotSupported;
switch (options.target.cpu.arch) {
.nvptx, .nvptx64 => {},
else => return error.PtxArchNotSupported,
}
if (!options.target.cpu.arch.isNvptx()) return error.PtxArchNotSupported;
switch (options.target.os.tag) {
// TODO: does it also work with nvcl ?
@ -59,9 +56,8 @@ pub fn openPath(allocator: Allocator, sub_path: []const u8, options: link.Option
if (!options.use_llvm) return error.PtxArchNotSupported;
assert(options.target.ofmt == .nvptx);
const nvptx = try createEmpty(allocator, options);
log.info("Opening .ptx target file {s}", .{sub_path});
return nvptx;
log.debug("Opening .ptx target file {s}", .{sub_path});
return createEmpty(allocator, options);
}
pub fn deinit(self: *NvPtx) void {
@ -76,15 +72,7 @@ pub fn updateFunc(self: *NvPtx, module: *Module, func: *Module.Fn, air: Air, liv
pub fn updateDecl(self: *NvPtx, module: *Module, decl_index: Module.Decl.Index) !void {
if (!build_options.have_llvm) return;
const decl = module.declPtr(decl_index);
log.info("updating {s}", .{decl.name});
return self.llvm_object.updateDecl(module, decl_index);
// const decl_index = func.owner_decl;
// const decl = module.declPtr(decl_index);
// try mod.decl_exports.ensureUnusedCapacity(gpa, 1);
// try mod.export_owners.ensureUnusedCapacity(gpa, 1);
// mod.decl_exports.getOrPutAssumeCapacity(exported_decl_index);
}
pub fn updateDeclExports(
@ -118,7 +106,7 @@ pub fn flushModule(self: *NvPtx, comp: *Compilation, prog_node: *std.Progress.No
defer tracy.end();
const outfile = comp.bin_file.options.emit.?;
// !!! We modify 'comp' before passing it to LLVM, but restore value afterwards
// We modify 'comp' before passing it to LLVM, but restore value afterwards.
// We tell LLVM to not try to build a .o, only an "assembly" file.
// This is required by the LLVM PTX backend.
comp.bin_file.options.emit = null;

View File

@ -411,13 +411,12 @@ pub fn classifyCompilerRtLibName(target: std.Target, name: []const u8) CompilerR
}
pub fn hasDebugInfo(target: std.Target) bool {
return switch (target.cpu.arch) {
.nvptx, .nvptx64 => {
// TODO: not sure to test "ptx >= 7.5" with featureset
return std.Target.nvptx.featureSetHas(target.cpu.features, .ptx75);
},
else => true
};
if (target.cpu.arch.isNvptx()) {
// TODO: not sure how to test "ptx >= 7.5" with featureset
return std.Target.nvptx.featureSetHas(target.cpu.features, .ptx75);
}
return true;
}
pub fn defaultCompilerRtOptimizeMode(target: std.Target) std.builtin.Mode {