Fix NVVMReflect pass usage. (#595)

Author: maleadt

.github/workflows/ci.yml

@@ -18,7 +18,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        version: ['1.8', '1.9', '1.10', '^1.11.0-rc1'] # 'nightly'
+        version: ['1.8', '1.9', '1.10', 'pre'] # 'nightly'
         os: [ubuntu-latest, macOS-latest, windows-latest]
         arch: [x64]
         llvm_args: ['']
@@ -36,15 +36,15 @@ jobs:
             os: 'windows-latest'
             arch: 'x64'
             llvm_args: '--opaque-pointers'
-          - version: '^1.11.0-rc1'
+          - version: 'pre'
             os: 'ubuntu-latest'
             arch: 'x64'
             llvm_args: '--opaque-pointers'
-          - version: '^1.11.0-rc1'
+          - version: 'pre'
             os: 'macOS-latest'
             arch: 'x64'
             llvm_args: '--opaque-pointers'
-          - version: '^1.11.0-rc1'
+          - version: 'pre'
             os: 'windows-latest'
             arch: 'x64'
             llvm_args: '--opaque-pointers'

src/irgen.jl

@@ -718,11 +718,10 @@ function add_kernel_state!(mod::LLVM.Module)
 
     return true
 end
+add_kernel_state!(pm::PassManager) =
+    add!(pm, ModulePass("AddKernelStatePass", add_kernel_state!))
 if LLVM.has_newpm()
     AddKernelStatePass() = NewPMModulePass("AddKernelStatePass", add_kernel_state!)
-else
-    add_kernel_state!(pm::PassManager) =
-        add!(pm, ModulePass("AddKernelStatePass", add_kernel_state!))
 end
 
 # lower calls to the state getter intrinsic. this is a two-step process, so that the state
@@ -775,11 +774,10 @@ function lower_kernel_state!(fun::LLVM.Function)
 
     return changed
 end
+lower_kernel_state!(pm::PassManager) =
+    add!(pm, FunctionPass("LowerKernelStatePass", lower_kernel_state!))
 if LLVM.has_newpm()
     LowerKernelStatePass() = NewPMFunctionPass("LowerKernelStatePass", lower_kernel_state!)
-else
-    lower_kernel_state!(pm::PassManager) =
-        add!(pm, FunctionPass("LowerKernelStatePass", lower_kernel_state!))
 end
 
 function cleanup_kernel_state!(mod::LLVM.Module)
@@ -798,11 +796,10 @@ function cleanup_kernel_state!(mod::LLVM.Module)
 
     return changed
 end
+cleanup_kernel_state!(pm::PassManager) =
+    add!(pm, ModulePass("CleanupKernelStatePass", cleanup_kernel_state!))
 if LLVM.has_newpm()
     CleanupKernelStatePass() = NewPMModulePass("CleanupKernelStatePass", cleanup_kernel_state!)
-else
-    cleanup_kernel_state!(pm::PassManager) =
-        add!(pm, ModulePass("CleanupKernelStatePass", cleanup_kernel_state!))
 end
 
 function kernel_state_intr(mod::LLVM.Module, T_state)

src/mcgen.jl

@@ -78,12 +78,11 @@ function resolve_cpu_references!(mod::LLVM.Module)
 
     return changed
 end
+resolve_cpu_references!(pm::PassManager) =
+    add!(pm, ModulePass("ResolveCPUReferences", resolve_cpu_references!))
 if LLVM.has_newpm()
     ResolveCPUReferencesPass() =
         NewPMModulePass("ResolveCPUReferences", resolve_cpu_references!)
-else
-    resolve_cpu_references!(pm::PassManager) =
-        add!(pm, ModulePass("ResolveCPUReferences", resolve_cpu_references!))
 end
 
 

src/optim.jl

@@ -648,10 +648,9 @@ function cpu_features!(mod::LLVM.Module)
 
     return changed
 end
+cpu_features!(pm::PassManager) = add!(pm, ModulePass("LowerCPUFeatures", cpu_features!))
 if LLVM.has_newpm()
     CPUFeaturesPass() = NewPMModulePass("GPULowerCPUFeatures", cpu_features!)
-else
-    cpu_features!(pm::PassManager) = add!(pm, ModulePass("LowerCPUFeatures", cpu_features!))
 end
 
 # lower object allocations to to PTX malloc
@@ -711,10 +710,9 @@ function lower_gc_frame!(fun::LLVM.Function)
 
     return changed
 end
+lower_gc_frame!(pm::PassManager) = add!(pm, FunctionPass("LowerGCFrame", lower_gc_frame!))
 if LLVM.has_newpm()
     LowerGCFramePass() = NewPMFunctionPass("GPULowerGCFrame", lower_gc_frame!)
-else
-    lower_gc_frame!(pm::PassManager) = add!(pm, FunctionPass("LowerGCFrame", lower_gc_frame!))
 end
 
 # lower the `julia.ptls_states` intrinsic by removing it, since it is GPU incompatible.
@@ -746,8 +744,7 @@ function lower_ptls!(mod::LLVM.Module)
 
     return changed
 end
+lower_ptls!(pm::PassManager) = add!(pm, ModulePass("LowerPTLS", lower_ptls!))
 if LLVM.has_newpm()
     LowerPTLSPass() = NewPMModulePass("GPULowerPTLS", lower_ptls!)
-else
-    lower_ptls!(pm::PassManager) = add!(pm, ModulePass("LowerPTLS", lower_ptls!))
 end

src/ptx.jl

@@ -156,6 +156,8 @@ function optimize_module!(@nospecialize(job::CompilerJob{PTXCompilerTarget}),
     # TODO: Use the registered target passes (JuliaGPU/GPUCompiler.jl#450)
     if use_newpm
         @dispose pb=NewPMPassBuilder() begin
+            register!(pb, NVVMReflectPass())
+
             add!(pb, NewPMFunctionPassManager()) do fpm
                 # TODO: need to run this earlier; optimize_module! is called after addOptimizationPasses!
                 add!(fpm, NVVMReflectPass())
@@ -523,9 +525,8 @@ function nvvm_reflect!(fun::LLVM.Function)
     end
     return changed
 end
+nvvm_reflect!(pm::PassManager) =
+    add!(pm, FunctionPass("NVVMReflect", nvvm_reflect!))
 if LLVM.has_newpm()
-    NVVMReflectPass() = NewPMFunctionPass("nvvm-reflect", nvvm_reflect!)
-else
-    nvvm_reflect!(pm::PassManager) =
-        add!(pm, FunctionPass("NVVMReflect", nvvm_reflect!))
+    NVVMReflectPass() = NewPMFunctionPass("custom-nvvm-reflect", nvvm_reflect!)
 end