This switches HIP from its currently convoluted macro + pfe based dispatch mechanism to a more natural one partially based on the existing module API. The basic idea is that HCC will always correctly emit __global__ functions: as empty-bodied stubs, on host, and as kernels, on device. It then becomes trivial to obtain the mangled name on host, at dispatch, from the function's address, and then to use the mangled name to retrieve the kernel. This should address all problems stemming from serialisation, dubious mismatches due to the manufactured functor, macro-isms et al. It also immediately enables support for generalised globals as a consequence of that being available in the module API. Finally, it will make debug much easier, since the actual names of the __global__ functions will automatically be used in traces etc. One detail is that due to how dispatch works now (hipLaunchKernel and hipLaunchKernelGGL are themselves variadic function templates which deduce the function type of the callee), in certain cases it may be necesssary to insert explicit casts to ensure that the variadic argument list selects a viable overload - this can be observed in some unit tests. Eventually we may be able to remove this limitation, but for now it does not appear terribly onerous. The code is not extremely HIPpie, nor is it fully optimised, but rather is intended as a starting point for the HIP team to make its own.

2017-11-01 15:09:59 +00:00
parent f27c2c1715
commit c2482d1255
27 changed files with 1457 additions and 1352 deletions
@@ -0,0 +1,134 @@
+#pragma once
+
+#include <hsa/hsa.h>
+
+#include <algorithm>
+#include <cstdint>
+#include <istream>
+#include <iterator>
+#include <string>
+#include <utility>
+#include <vector>
+
+namespace hip_impl
+{
+    hsa_isa_t triple_to_hsa_isa(const std::string& triple);
+
+    struct Bundled_code {
+        union {
+            struct {
+                std::uint64_t offset;
+                std::uint64_t bundle_sz;
+                std::uint64_t triple_sz;
+            };
+            std::uint8_t cbuf[
+                sizeof(offset) + sizeof(bundle_sz) + sizeof(triple_sz)];
+        };
+        std::string triple;
+        std::vector<std::uint8_t> blob;
+    };
+
+    class Bundled_code_header {
+        // DATA - STATICS
+        static constexpr const char magic_string_[] =
+            "__CLANG_OFFLOAD_BUNDLE__";
+        static constexpr auto magic_string_sz_ = sizeof(magic_string_) - 1;
+
+        // DATA
+        union {
+            struct {
+                std::uint8_t bundler_magic_string_[magic_string_sz_];
+                std::uint64_t bundle_cnt_;
+            };
+            std::uint8_t cbuf_[
+                sizeof(bundler_magic_string_) + sizeof(bundle_cnt_)];
+        };
+        std::vector<Bundled_code> bundles_;
+
+        // FRIENDS - MANIPULATORS
+        template<typename RandomAccessIterator>
+        friend
+        inline
+        bool read(
+            RandomAccessIterator f,
+            RandomAccessIterator l,
+            Bundled_code_header& x)
+        {
+            std::copy_n(f, sizeof(x.cbuf_), x.cbuf_);
+
+            if (valid(x)) {
+                x.bundles_.resize(x.bundle_cnt_);
+
+                auto it = f + sizeof(x.cbuf_);
+                for (auto&& y : x.bundles_) {
+                    std::copy_n(it, sizeof(y.cbuf), y.cbuf);
+                    it += sizeof(y.cbuf);
+
+                    y.triple.insert(y.triple.cend(), it, it + y.triple_sz);
+
+                    std::copy_n(
+                        f + y.offset, y.bundle_sz, std::back_inserter(y.blob));
+
+                    it += y.triple_sz;
+                }
+
+                return true;
+            }
+
+            return false;
+        }
+        friend
+        inline
+        bool read(const std::vector<std::uint8_t>& blob, Bundled_code_header& x)
+        {
+            return read(blob.cbegin(), blob.cend(), x);
+        }
+        friend
+        inline
+        bool read(std::istream& is, Bundled_code_header& x)
+        {
+            return read(std::vector<std::uint8_t>{
+                std::istreambuf_iterator<char>{is},
+                std::istreambuf_iterator<char>{}},
+                x);
+        }
+
+        // FRIENDS - ACCESSORS
+        friend
+        inline
+        bool valid(const Bundled_code_header& x)
+        {
+            return std::equal(
+                x.bundler_magic_string_,
+                x.bundler_magic_string_ + magic_string_sz_,
+                x.magic_string_);
+        }
+        friend
+        inline
+        const std::vector<Bundled_code>& bundles(const Bundled_code_header& x)
+        {
+            return x.bundles_;
+        }
+    public:
+        // CREATORS
+        Bundled_code_header() = default;
+        template<typename RandomAccessIterator>
+        Bundled_code_header(RandomAccessIterator f, RandomAccessIterator l);
+        explicit
+        Bundled_code_header(const std::vector<std::uint8_t>& blob);
+        Bundled_code_header(const Bundled_code_header&) = default;
+        Bundled_code_header(Bundled_code_header&&) = default;
+        ~Bundled_code_header() = default;
+
+        // MANIPULATORS
+        Bundled_code_header& operator=(const Bundled_code_header&) = default;
+        Bundled_code_header& operator=(Bundled_code_header&&) = default;
+    };
+
+    // CREATORS
+    template<typename I>
+    Bundled_code_header::Bundled_code_header(I f, I l) : Bundled_code_header{}
+    {
+        read(f, l, *this);
+    }
+} // Namespace hip_impl.