Refined PTX argument mapping and analyses.

.github/workflows/check-required.yml

@@ -0,0 +1,60 @@
+name: Check required jobs
+
+# This workflow is triggered when a workflow run for the CI is completed.
+# It checks if the "All required checks done" job was actually successful
+# (and not just skipped) and creates a check run if that is the case. The
+# check run can be used to protect the main branch from being merged if the
+# CI is not passing.
+
+on:
+  workflow_run:
+    types: [completed]
+    workflows:
+      - CI
+      - Deploy Site with Jekyll, GitHub Pages
+
+permissions:
+  actions: read
+  checks: write
+
+jobs:
+  required-jobs:
+    name: Check required jobs
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/github-script@v6
+        with:
+          script: |
+            // list jobs for worklow run attempt
+            const { data: { jobs } } = await github.rest.actions.listJobsForWorkflowRunAttempt({
+              owner: context.payload.repository.owner.login,
+              repo: context.payload.repository.name,
+              run_id: context.payload.workflow_run.id,
+              attempt_number: context.payload.workflow_run.run_attempt,
+            });
+            // check if required job was successful
+            var success = false;
+            core.info(`Checking jobs for workflow run ${context.payload.workflow_run.html_url}`);
+            jobs.forEach(job => {
+              var mark = '-'
+              if (job.name === 'All required checks done' && job.conclusion === 'success') {
+                success = true;
+                mark = '✅';
+              }
+              core.info(`${mark} ${job.name}: ${job.conclusion}`);
+            });
+            // create check run if job was successful
+            if (success) {
+              await github.rest.checks.create({
+                owner: context.payload.repository.owner.login,
+                repo: context.payload.repository.name,
+                name: 'All required checks succeeded',
+                head_sha: context.payload.workflow_run.head_sha,
+                status: 'completed',
+                conclusion: 'success',
+                output: {
+                  title: 'All required checks succeeded',
+                  summary: `See [workflow run](${context.payload.workflow_run.html_url}) for details.`,
+                },
+              });
+            }

.github/workflows/ci.yml

@@ -116,7 +116,7 @@ jobs:
             )
           ) && os+=("macos-latest")
 
-          [ "${{ steps.is-fork.outputs.fork }}" == "false" ] && os+=("cuda-${{ github.run_id }}-${{ github.run_attempt }}")
+          [ "${{ steps.is-fork.outputs.fork }}" == "false" ] && os+=("cuda")
 
           echo "os=$(jq -cn '$ARGS.positional' --args ${os[@]})" >> $GITHUB_OUTPUT
     outputs:
@@ -162,46 +162,38 @@ jobs:
             library: ILGPU.Algorithms
             framework: net471
       fail-fast: false
-    runs-on: ${{ contains(matrix.os, 'cuda') && format('{0}-{1}-{2}', matrix.os, matrix.library, matrix.framework) || matrix.os }}
+    runs-on: ${{ matrix.os }}
     steps:
       - name: Checkout
         uses: actions/checkout@v3
 
       - name: Setup the latest .NET Core 3.1 SDK
         if: matrix.framework == 'netcoreapp3.1'
         uses: actions/setup-dotnet@v3.2.0
-        env:
-          DOTNET_INSTALL_DIR: ${{ matrix.framework == 'cuda' && '~/.' }}
         with:
           dotnet-version: 3.1.x
 
       - name: Setup the latest .NET 5 SDK
         if: matrix.framework == 'net5.0'
         uses: actions/setup-dotnet@v3.2.0
-        env:
-          DOTNET_INSTALL_DIR: ${{ matrix.framework == 'cuda' && '~/.' }}
         with:
           dotnet-version: 5.0.x
 
       - name: Setup the latest .NET 6 SDK
         if: matrix.framework == 'net6.0'
         uses: actions/setup-dotnet@v3.2.0
-        env:
-          DOTNET_INSTALL_DIR: ${{ matrix.framework == 'cuda' && '~/.' }}
         with:
           dotnet-version: 6.0.x
 
       - name: Setup the latest .NET 7 SDK
         uses: actions/setup-dotnet@v3.2.0
-        env:
-          DOTNET_INSTALL_DIR: ${{ matrix.framework == 'cuda' && '~/.' }}
         with:
           dotnet-version: 7.0.x
 
       - name: Set test flavor
         id: test-flavor
         shell: bash
-        run: echo "flavor=$([[ "${{ matrix.os }}" == cuda-* ]] && echo "Cuda" || echo "CPU")" >> $GITHUB_OUTPUT
+        run: echo "flavor=$([[ "${{ matrix.os }}" == cuda ]] && echo "Cuda" || echo "CPU")" >> $GITHUB_OUTPUT
 
       - name: Build and test
         run: dotnet test Src/${{ matrix.library }}.Tests.${{ steps.test-flavor.outputs.flavor }} --configuration=Release --framework=${{ matrix.framework }} -p:TreatWarningsAsErrors=true
@@ -341,6 +333,9 @@ jobs:
 
   # Virtual job that can be configured as a required check before a PR can be
   # merged.
+  # As GitHub considers a check as successful if it is skipped, we need to
+  # check its status in another workflow (check-required.yml) and create a
+  # check there.
   all-required-checks-done:
     needs:
       - check-style

.github/workflows/deploy-site.yml

@@ -77,6 +77,11 @@ jobs:
         uses: actions/upload-pages-artifact@v2
       ##### GitHub Pages #####
 
+  # Virtual job that can be configured as a required check before a PR can be
+  # merged.
+  # As GitHub considers a check as successful if it is skipped, we need to
+  # check its status in another workflow (check-required.yml) and create a
+  # check there.
   all-required-checks-done:
     needs:
       - build

Src/ILGPU/Backends/OpenCL/CLCodeGenerator.Emitter.cs

@@ -47,6 +47,7 @@ public struct StatementEmitter : IDisposable
             #region Instance
 
             private readonly StringBuilder stringBuilder;
+            private readonly StringBuilder variableBuilder;
             private bool argMode;
             private int argumentCount;
 
@@ -58,6 +59,7 @@ internal StatementEmitter(CLCodeGenerator codeGenerator)
             {
                 CodeGenerator = codeGenerator;
                 stringBuilder = codeGenerator.Builder;
+                variableBuilder = codeGenerator.VariableBuilder;
                 argumentCount = 0;
                 argMode = false;
 
@@ -85,20 +87,23 @@ internal StatementEmitter(CLCodeGenerator codeGenerator)
             private void BeginAppendTarget(Variable target, bool appendNew = true)
             {
                 if (appendNew)
-                {
-                    var variableType = CodeGenerator.GetVariableType(target);
-                    stringBuilder.Append(variableType);
-                    stringBuilder.Append(' ');
-                }
+                    AppendDeclaration(target);
                 stringBuilder.Append(target.ToString());
             }
 
             /// <summary>
             /// Appends a target declaration.
             /// </summary>
             /// <param name="target">The target declaration.</param>
-            internal void AppendDeclaration(Variable target) =>
-                BeginAppendTarget(target);
+            internal void AppendDeclaration(Variable target)
+            {
+                var variableType = CodeGenerator.GetVariableType(target);
+                variableBuilder.Append('\t');
+                variableBuilder.Append(variableType);
+                variableBuilder.Append(' ');
+                variableBuilder.Append(target.ToString());
+                variableBuilder.AppendLine(";");
+            }
 
             /// <summary>
             /// Appends a target.
@@ -828,6 +833,50 @@ public StatementEmitter BeginStatement(FormattableString command)
             return emitter;
         }
 
+        /// <summary>
+        /// Begins the function body, switching to variable capturing mode.
+        /// </summary>
+        protected void BeginFunctionBody()
+        {
+            // Start the function body.
+            Builder.AppendLine("{");
+            PushIndent();
+
+#if DEBUG
+            Builder.AppendLine();
+            Builder.AppendLine("\t// Variable declarations");
+            Builder.AppendLine();
+#endif
+
+            // Switch to the alternate builder, so that we can capture the code and
+            // variable declarations separately.
+            prefixBuilder = Builder;
+            Builder = suffixBuilder;
+        }
+
+        /// <summary>
+        /// Finishes the function body, ending variable capturing mode.
+        /// </summary>
+        protected void FinishFunctionBody()
+        {
+            // Restore the original builder, containing code before the variable
+            // declarations.
+            Builder = prefixBuilder;
+
+            // Add the variable declarations at the start of the function, to avoid
+            // issues with OpenCL compilers that are not C99 compliant, and cannot
+            // handle variable declarations intermingled with other code.
+            Builder.Append(VariableBuilder);
+            Builder.AppendLine();
+
+            // Add the code that was generated along with the variable declarations.
+            Builder.Append(suffixBuilder);
+
+            // Close the function body.
+            PopIndent();
+            Builder.AppendLine("}");
+        }
+
         #endregion
     }
 }

Src/ILGPU/Backends/OpenCL/CLCodeGenerator.cs

@@ -279,6 +279,9 @@ protected static string GetParameterName(Parameter parameter) =>
             new Dictionary<BasicBlock, string>();
         private readonly string labelPrefix;
 
+        private StringBuilder prefixBuilder = new StringBuilder();
+        private StringBuilder suffixBuilder = new StringBuilder();
+
         /// <summary>
         /// Constructs a new code generator.
         /// </summary>
@@ -295,7 +298,7 @@ internal CLCodeGenerator(in GeneratorArgs args, Method method, Allocas allocas)
 
             labelPrefix = "L_" + Method.Id.ToString();
 
-            Builder = new StringBuilder();
+            Builder = prefixBuilder;
         }
 
         #endregion
@@ -327,7 +330,12 @@ public IntrinsicImplementationProvider<CLIntrinsic.Handler>
         /// <summary>
         /// Returns the associated string builder.
         /// </summary>
-        public StringBuilder Builder { get; }
+        public StringBuilder Builder { get; private set; }
+
+        /// <summary>
+        /// Returns the associated string builder.
+        /// </summary>
+        public StringBuilder VariableBuilder { get; } = new StringBuilder();
 
         #endregion
 

Src/ILGPU/Backends/OpenCL/CLFunctionGenerator.cs

@@ -132,11 +132,9 @@ public override void GenerateCode()
             BindSharedMemoryAllocation(Allocas.DynamicSharedAllocations);
 
             // Generate code
-            Builder.AppendLine("{");
-            PushIndent();
+            BeginFunctionBody();
             GenerateCodeInternal();
-            PopIndent();
-            Builder.AppendLine("}");
+            FinishFunctionBody();
         }
 
         #endregion

Src/ILGPU/Backends/OpenCL/CLKernelFunctionGenerator.cs

@@ -262,8 +262,7 @@ public override void GenerateCode()
             Builder.AppendLine(")");
 
             // Emit code that moves view arguments into their appropriate targets
-            Builder.AppendLine("{");
-            PushIndent();
+            BeginFunctionBody();
             GenerateArgumentMapping();
 
             // Emit index computation
@@ -323,8 +322,7 @@ public override void GenerateCode()
 
             // Generate code
             GenerateCodeInternal();
-            PopIndent();
-            Builder.AppendLine("}");
+            FinishFunctionBody();
         }
 
         /// <summary>

Src/ILGPU/Backends/PTX/Analyses/DefaultPTXBlockSchedule.cs

@@ -1,6 +1,6 @@
 // ---------------------------------------------------------------------------------------
 //                                        ILGPU
-//                        Copyright (c) 2020-2021 ILGPU Project
+//                        Copyright (c) 2020-2023 ILGPU Project
 //                                    www.ilgpu.net
 //
 // File: DefaultPTXBlockSchedule.cs
@@ -18,7 +18,7 @@ namespace ILGPU.Backends.PTX.Analyses
     /// <summary>
     /// Represents a default PTX-specific block schedule.
     /// </summary>
-    public sealed class DefaultPTXBlockSchedule :
+    sealed class DefaultPTXBlockSchedule :
         PTXBlockSchedule<ReversePostOrder, Forwards>
     {
         #region Instance
@@ -53,7 +53,7 @@ public override bool IsImplicitSuccessor(
         #endregion
     }
 
-    public partial class PTXBlockScheduleExtensions
+    partial class PTXBlockScheduleExtensions
     {
         /// <summary>
         /// Creates a new default block schedule using the given blocks.