Merge topic 'support_cubin_fatbin_optix_cuda_output'

2def6a874b CUDA: Add support for CUBIN, FATBIN, and OPTIXIR compilation

Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: buildbot <buildbot@kitware.com>
Merge-request: !8259

5 files changed, 128 insertions, 34 deletions

diff --git a/Source/cmGeneratorTarget.cxx b/Source/cmGeneratorTarget.cxx
index 54cd8ba..4cfa1d7 100644
--- a/Source/cmGeneratorTarget.cxx
+++ b/Source/cmGeneratorTarget.cxx
@@ -3,6 +3,7 @@
 #include "cmGeneratorTarget.h"
 
 #include <algorithm>
+#include <array>
 #include <cassert>
 #include <cerrno>
 #include <cstddef>
@@ -1011,12 +1012,27 @@ const std::string& cmGeneratorTarget::GetObjectName(cmSourceFile const* file)
 
 const char* cmGeneratorTarget::GetCustomObjectExtension() const
 {
-  static std::string extension;
-  const bool has_ptx_extension =
-    this->GetPropertyAsBool("CUDA_PTX_COMPILATION");
-  if (has_ptx_extension) {
-    extension = ".ptx";
-    return extension.c_str();
+  struct compiler_mode
+  {
+    std::string variable;
+    std::string extension;
+  };
+  static std::array<compiler_mode, 4> const modes{
+    { { "CUDA_PTX_COMPILATION", ".ptx" },
+      { "CUDA_CUBIN_COMPILATION", ".cubin" },
+      { "CUDA_FATBIN_COMPILATION", ".fatbin" },
+      { "CUDA_OPTIX_COMPILATION", ".optixir" } }
+  };
+
+  std::string const& compiler =
+    this->Makefile->GetSafeDefinition("CMAKE_CUDA_COMPILER_ID");
+  if (!compiler.empty()) {
+    for (const auto& m : modes) {
+      const bool has_extension = this->GetPropertyAsBool(m.variable);
+      if (has_extension) {
+        return m.extension.c_str();
+      }
+    }
   }
   return nullptr;
 }
diff --git a/Source/cmMakefileTargetGenerator.cxx b/Source/cmMakefileTargetGenerator.cxx
index e217dd9..2ead739 100644
--- a/Source/cmMakefileTargetGenerator.cxx
+++ b/Source/cmMakefileTargetGenerator.cxx
@@ -3,6 +3,7 @@
 #include "cmMakefileTargetGenerator.h"
 
 #include <algorithm>
+#include <array>
 #include <cassert>
 #include <cstdio>
 #include <iterator>
@@ -977,11 +978,23 @@ void cmMakefileTargetGenerator::WriteObjectRuleFiles(
           this->Makefile->GetRequiredDefinition("_CMAKE_CUDA_RDC_FLAG");
         cudaCompileMode = cmStrCat(cudaCompileMode, rdcFlag, " ");
       }
-      if (this->GeneratorTarget->GetPropertyAsBool("CUDA_PTX_COMPILATION")) {
-        const std::string& ptxFlag =
-          this->Makefile->GetRequiredDefinition("_CMAKE_CUDA_PTX_FLAG");
-        cudaCompileMode = cmStrCat(cudaCompileMode, ptxFlag);
-      } else {
+
+      static std::array<cm::string_view, 4> const compileModes{
+        { "PTX"_s, "CUBIN"_s, "FATBIN"_s, "OPTIX"_s }
+      };
+      bool useNormalCompileMode = true;
+      for (cm::string_view mode : compileModes) {
+        auto propName = cmStrCat("CUDA_", mode, "_COMPILATION");
+        auto defName = cmStrCat("_CMAKE_CUDA_", mode, "_FLAG");
+        if (this->GeneratorTarget->GetPropertyAsBool(propName)) {
+          const std::string& flag =
+            this->Makefile->GetRequiredDefinition(defName);
+          cudaCompileMode = cmStrCat(cudaCompileMode, flag);
+          useNormalCompileMode = false;
+          break;
+        }
+      }
+      if (useNormalCompileMode) {
         const std::string& wholeFlag =
           this->Makefile->GetRequiredDefinition("_CMAKE_CUDA_WHOLE_FLAG");
         cudaCompileMode = cmStrCat(cudaCompileMode, wholeFlag);
diff --git a/Source/cmNinjaTargetGenerator.cxx b/Source/cmNinjaTargetGenerator.cxx
index 8663f46..dc56142 100644
--- a/Source/cmNinjaTargetGenerator.cxx
+++ b/Source/cmNinjaTargetGenerator.cxx
@@ -3,6 +3,7 @@
 #include "cmNinjaTargetGenerator.h"
 
 #include <algorithm>
+#include <array>
 #include <cassert>
 #include <functional>
 #include <iterator>
@@ -859,11 +860,22 @@ void cmNinjaTargetGenerator::WriteCompileRule(const std::string& lang,
         this->Makefile->GetRequiredDefinition("_CMAKE_CUDA_RDC_FLAG");
       cudaCompileMode = cmStrCat(cudaCompileMode, rdcFlag, " ");
     }
-    if (this->GeneratorTarget->GetPropertyAsBool("CUDA_PTX_COMPILATION")) {
-      const std::string& ptxFlag =
-        this->Makefile->GetRequiredDefinition("_CMAKE_CUDA_PTX_FLAG");
-      cudaCompileMode = cmStrCat(cudaCompileMode, ptxFlag);
-    } else {
+    static std::array<cm::string_view, 4> const compileModes{
+      { "PTX"_s, "CUBIN"_s, "FATBIN"_s, "OPTIX"_s }
+    };
+    bool useNormalCompileMode = true;
+    for (cm::string_view mode : compileModes) {
+      auto propName = cmStrCat("CUDA_", mode, "_COMPILATION");
+      auto defName = cmStrCat("_CMAKE_CUDA_", mode, "_FLAG");
+      if (this->GeneratorTarget->GetPropertyAsBool(propName)) {
+        const std::string& flag =
+          this->Makefile->GetRequiredDefinition(defName);
+        cudaCompileMode = cmStrCat(cudaCompileMode, flag);
+        useNormalCompileMode = false;
+        break;
+      }
+    }
+    if (useNormalCompileMode) {
       const std::string& wholeFlag =
         this->Makefile->GetRequiredDefinition("_CMAKE_CUDA_WHOLE_FLAG");
       cudaCompileMode = cmStrCat(cudaCompileMode, wholeFlag);
@@ -1789,11 +1801,22 @@ void cmNinjaTargetGenerator::ExportObjectCompileCommand(
         this->Makefile->GetRequiredDefinition("_CMAKE_CUDA_RDC_FLAG");
       cudaCompileMode = cmStrCat(cudaCompileMode, rdcFlag, " ");
     }
-    if (this->GeneratorTarget->GetPropertyAsBool("CUDA_PTX_COMPILATION")) {
-      const std::string& ptxFlag =
-        this->Makefile->GetRequiredDefinition("_CMAKE_CUDA_PTX_FLAG");
-      cudaCompileMode = cmStrCat(cudaCompileMode, ptxFlag);
-    } else {
+    static std::array<cm::string_view, 4> const compileModes{
+      { "PTX"_s, "CUBIN"_s, "FATBIN"_s, "OPTIX"_s }
+    };
+    bool useNormalCompileMode = true;
+    for (cm::string_view mode : compileModes) {
+      auto propName = cmStrCat("CUDA_", mode, "_COMPILATION");
+      auto defName = cmStrCat("_CMAKE_CUDA_", mode, "_FLAG");
+      if (this->GeneratorTarget->GetPropertyAsBool(propName)) {
+        const std::string& flag =
+          this->Makefile->GetRequiredDefinition(defName);
+        cudaCompileMode = cmStrCat(cudaCompileMode, flag);
+        useNormalCompileMode = false;
+        break;
+      }
+    }
+    if (useNormalCompileMode) {
       const std::string& wholeFlag =
         this->Makefile->GetRequiredDefinition("_CMAKE_CUDA_WHOLE_FLAG");
       cudaCompileMode = cmStrCat(cudaCompileMode, wholeFlag);
diff --git a/Source/cmTarget.cxx b/Source/cmTarget.cxx
index ec87271..f06e26b 100644
--- a/Source/cmTarget.cxx
+++ b/Source/cmTarget.cxx
@@ -1775,6 +1775,9 @@ MAKE_PROP(COMPILE_FEATURES);
 MAKE_PROP(COMPILE_OPTIONS);
 MAKE_PROP(PRECOMPILE_HEADERS);
 MAKE_PROP(PRECOMPILE_HEADERS_REUSE_FROM);
+MAKE_PROP(CUDA_CUBIN_COMPILATION);
+MAKE_PROP(CUDA_FATBIN_COMPILATION);
+MAKE_PROP(CUDA_OPTIX_COMPILATION);
 MAKE_PROP(CUDA_PTX_COMPILATION);
 MAKE_PROP(EXPORT_NAME);
 MAKE_PROP(IMPORTED);
@@ -1910,14 +1913,38 @@ void cmTarget::StoreProperty(const std::string& prop, ValueType value)
                value ? value
                      : std::string{})) { // NOLINT(bugprone-branch-clone)
     /* error was reported by check method */
-  } else if (prop == propCUDA_PTX_COMPILATION &&
-             this->GetType() != cmStateEnums::OBJECT_LIBRARY) {
-    std::ostringstream e;
-    e << "CUDA_PTX_COMPILATION property can only be applied to OBJECT "
-         "targets (\""
-      << this->impl->Name << "\")\n";
-    this->impl->Makefile->IssueMessage(MessageType::FATAL_ERROR, e.str());
-    return;
+  } else if (prop == propCUDA_CUBIN_COMPILATION ||
+             prop == propCUDA_FATBIN_COMPILATION ||
+             prop == propCUDA_OPTIX_COMPILATION ||
+             prop == propCUDA_PTX_COMPILATION) {
+    auto const& compiler =
+      this->impl->Makefile->GetSafeDefinition("CMAKE_CUDA_COMPILER_ID");
+    auto const& compilerVersion =
+      this->impl->Makefile->GetSafeDefinition("CMAKE_CUDA_COMPILER_VERSION");
+    if (this->GetType() != cmStateEnums::OBJECT_LIBRARY) {
+      auto e =
+        cmStrCat(prop, " property can only be applied to OBJECT targets(",
+                 this->impl->Name, ")\n");
+      this->impl->Makefile->IssueMessage(MessageType::FATAL_ERROR, e);
+      return;
+    }
+    const bool flag_found =
+      (prop == propCUDA_PTX_COMPILATION &&
+       this->impl->Makefile->GetDefinition("_CMAKE_CUDA_PTX_FLAG")) ||
+      (prop == propCUDA_CUBIN_COMPILATION &&
+       this->impl->Makefile->GetDefinition("_CMAKE_CUDA_CUBIN_FLAG")) ||
+      (prop == propCUDA_FATBIN_COMPILATION &&
+       this->impl->Makefile->GetDefinition("_CMAKE_CUDA_FATBIN_FLAG")) ||
+      (prop == propCUDA_OPTIX_COMPILATION &&
+       this->impl->Makefile->GetDefinition("_CMAKE_CUDA_OPTIX_FLAG"));
+    if (flag_found) {
+      this->impl->Properties.SetProperty(prop, value);
+    } else {
+      auto e = cmStrCat(prop, " property is not supported by ", compiler,
+                        "  compiler version ", compilerVersion, ".");
+      this->impl->Makefile->IssueMessage(MessageType::FATAL_ERROR, e);
+      return;
+    }
   } else if (prop == propPRECOMPILE_HEADERS_REUSE_FROM) {
     if (this->GetProperty("PRECOMPILE_HEADERS")) {
       std::ostringstream e;
diff --git a/Source/cmVisualStudio10TargetGenerator.cxx b/Source/cmVisualStudio10TargetGenerator.cxx
index ef0fcf3..8926f9e 100644
--- a/Source/cmVisualStudio10TargetGenerator.cxx
+++ b/Source/cmVisualStudio10TargetGenerator.cxx
@@ -3597,13 +3597,13 @@ bool cmVisualStudio10TargetGenerator::ComputeCudaOptions(
   if (this->GeneratorTarget->GetPropertyAsBool("CUDA_SEPARABLE_COMPILATION")) {
     cudaOptions.AddFlag("GenerateRelocatableDeviceCode", "true");
   }
-  bool notPtx = true;
+  bool notPtxLike = true;
   if (this->GeneratorTarget->GetPropertyAsBool("CUDA_PTX_COMPILATION")) {
     cudaOptions.AddFlag("NvccCompilation", "ptx");
     // We drop the %(Extension) component as CMake expects all PTX files
     // to not have the source file extension at all
     cudaOptions.AddFlag("CompileOut", "$(IntDir)%(Filename).ptx");
-    notPtx = false;
+    notPtxLike = false;
 
     if (cmSystemTools::VersionCompare(cmSystemTools::OP_GREATER_EQUAL,
                                       cudaVersion, "9.0") &&
@@ -3618,9 +3618,24 @@ bool cmVisualStudio10TargetGenerator::ComputeCudaOptions(
                           "%(BaseCommandLineTemplate) [CompileOut] [FastMath] "
                           "[Defines] \"%(FullPath)\"");
     }
-  }
-
-  if (notPtx &&
+  } else if (this->GeneratorTarget->GetPropertyAsBool(
+               "CUDA_CUBIN_COMPILATION")) {
+    cudaOptions.AddFlag("NvccCompilation", "cubin");
+    cudaOptions.AddFlag("CompileOut", "$(IntDir)%(Filename).cubin");
+    notPtxLike = false;
+  } else if (this->GeneratorTarget->GetPropertyAsBool(
+               "CUDA_FATBIN_COMPILATION")) {
+    cudaOptions.AddFlag("NvccCompilation", "fatbin");
+    cudaOptions.AddFlag("CompileOut", "$(IntDir)%(Filename).fatbin");
+    notPtxLike = false;
+  } else if (this->GeneratorTarget->GetPropertyAsBool(
+               "CUDA_OPTIX_COMPILATION")) {
+    cudaOptions.AddFlag("NvccCompilation", "optix-ir");
+    cudaOptions.AddFlag("CompileOut", "$(IntDir)%(Filename).optixir");
+    notPtxLike = false;
+  }
+
+  if (notPtxLike &&
       cmSystemTools::VersionCompareGreaterEq(
         "8.0", this->GlobalGenerator->GetPlatformToolsetCudaString())) {
     // Explicitly state that we want this file to be treated as a