runner

Signed-off-by: Larry Gritz <lg@larrygritz.com>

.github/workflows/analysis.yml

@@ -19,6 +19,11 @@ on:
 
 permissions: read-all
 
+# Allow subsequent pushes to the same PR or REF to cancel any previous jobs.
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
 
 jobs:
 
@@ -30,9 +35,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - desc: sonary gcc9/C++14 llvm13 py3.9 boost1.76 exr3.1 oiio2.3 avx2
+          - desc: sonar gcc9/C++14 llvm13 py3.9 boost1.76 exr3.1 oiio2.3 avx2
             nametag: static-analysis-sonar
-            os: ubuntu-latest
+            os: ubuntu-20.04-8c-32g-300h
+            # os: ubuntu-latest
             vfxyear: 2022
             vfxsuffix: -clang13
             cxx_std: 14
@@ -40,7 +46,7 @@ jobs:
             python_ver: 3.9
             pybind11_ver: v2.9.0
             simd: avx2,f16c
-            batched: b8_AVX2
+            batched: b8_AVX2,b16_AVX512
             coverage: 1
             # skip_tests: 1
             sonar: 1
@@ -71,6 +77,7 @@ jobs:
       PYTHON_VERSION: ${{matrix.python_ver}}
       USE_BATCHED: ${{matrix.batched}}
       USE_SIMD: ${{matrix.simd}}
+      PARALLEL: 24
       # DEBUG_CI: 1
     steps:
       # We would like to use harden-runner, but it flags too many false

.github/workflows/ci.yml

@@ -25,6 +25,11 @@ on:
 
 permissions: read-all
 
+# Allow subsequent pushes to the same PR or REF to cancel any previous jobs.
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
 
 jobs:
 

src/build-scripts/ci-startup.bash

@@ -61,6 +61,7 @@ echo "uname -a: " `uname -a`
 echo "uname -m: " `uname -m`
 echo "uname -s: " `uname -s`
 echo "uname -n: " `uname -n`
+echo "nproc " `nproc`
 pwd
 ls
 env | sort

src/include/OSL/llvm_util.h

@@ -154,15 +154,6 @@ public:
     void debug_pop_inlined_function();
     void debug_set_location(OIIO::ustring sourcefile, int sourceline);
 
-    /// Create a new function (that will later be populated with
-    /// instructions) with up to 4 args.
-    OSL_DEPRECATED("Use make_function flavor that takes a cspan (1.12)")
-    llvm::Function* make_function(const std::string& name, bool fastcall,
-                                  llvm::Type* rettype, llvm::Type* arg1 = NULL,
-                                  llvm::Type* arg2 = NULL,
-                                  llvm::Type* arg3 = NULL,
-                                  llvm::Type* arg4 = NULL);
-
     /// Create a new function (that will later be populated with
     /// instructions) with a cspan of args.
     llvm::Function* make_function(const std::string& name, bool fastcall,

src/liboslexec/llvm_util.cpp

@@ -1589,29 +1589,26 @@ LLVM_Util::make_jit_execengine(std::string* err, TargetISA requestedISA,
 
 
 
-namespace /*anonymous*/ {
+#ifdef OSL_DEV
 // The return value of llvm::StructLayout::getAlignment()
 // changed from an int to llvm::Align, hide with accessor function
-#if OSL_LLVM_VERSION < 100
-uint64_t
+inline uint64_t
 get_alignment(const llvm::StructLayout* layout)
 {
+#    if OSL_LLVM_VERSION < 100
     return layout->getAlignment();
-}
-#else
-uint64_t
-get_alignment(const llvm::StructLayout* layout)
-{
+#    else
     return layout->getAlignment().value();
+#    endif
 }
 #endif
-}  // namespace
 
 
 
 void
-LLVM_Util::dump_struct_data_layout(llvm::Type* Ty)
+LLVM_Util::dump_struct_data_layout(llvm::Type* Ty OSL_MAYBE_UNUSED)
 {
+#ifdef OSL_DEV
     OSL_ASSERT(Ty);
     OSL_ASSERT(Ty->isStructTy());
 
@@ -1635,6 +1632,7 @@ LLVM_Util::dump_struct_data_layout(llvm::Type* Ty)
         }
         std::cout << std::endl;
     }
+#endif
 }
 
 
@@ -2449,25 +2447,6 @@ LLVM_Util::internalize_module_functions(
 
 
 
-llvm::Function*
-LLVM_Util::make_function(const std::string& name, bool fastcall,
-                         llvm::Type* rettype, llvm::Type* arg1,
-                         llvm::Type* arg2, llvm::Type* arg3, llvm::Type* arg4)
-{
-    std::vector<llvm::Type*> argtypes;
-    if (arg1)
-        argtypes.emplace_back(arg1);
-    if (arg2)
-        argtypes.emplace_back(arg2);
-    if (arg3)
-        argtypes.emplace_back(arg3);
-    if (arg4)
-        argtypes.emplace_back(arg4);
-    return make_function(name, fastcall, rettype, argtypes, false);
-}
-
-
-
 llvm::Function*
 LLVM_Util::make_function(const std::string& name, bool fastcall,
                          llvm::Type* rettype, cspan<llvm::Type*> params,
@@ -5831,6 +5810,7 @@ bool
 LLVM_Util::ptx_compile_group(llvm::Module* lib_module, const std::string& name,
                              std::string& out)
 {
+#ifdef OSL_USE_OPTIX
     std::string target_triple = module()->getTargetTriple();
 
     OSL_ASSERT(
@@ -5874,9 +5854,9 @@ LLVM_Util::ptx_compile_group(llvm::Module* lib_module, const std::string& name,
     options.AllowFPOpFusion                        = llvm::FPOpFusion::Fast;
     options.NoZerosInBSS                           = 0;
     options.GuaranteedTailCallOpt                  = 0;
-#if OSL_LLVM_VERSION < 120
+#    if OSL_LLVM_VERSION < 120
     options.StackAlignmentOverride = 0;
-#endif
+#    endif
     options.UseInitArray = 0;
 
     llvm::TargetMachine* target_machine = llvm_target->createTargetMachine(
@@ -5908,15 +5888,15 @@ LLVM_Util::ptx_compile_group(llvm::Module* lib_module, const std::string& name,
     llvm::raw_svector_ostream assembly_stream(assembly);
 
     // TODO: Make sure rounding modes, etc., are set correctly
-#if OSL_LLVM_VERSION >= 100
+#    if OSL_LLVM_VERSION >= 100
     target_machine->addPassesToEmitFile(mod_pm, assembly_stream,
                                         nullptr,  // FIXME: Correct?
                                         llvm::CGFT_AssemblyFile);
-#else
+#    else
     target_machine->addPassesToEmitFile(mod_pm, assembly_stream,
                                         nullptr,  // FIXME: Correct?
                                         llvm::TargetMachine::CGFT_AssemblyFile);
-#endif
+#    endif
 
     // Run the optimization passes on the functions
     fn_pm.doInitialization();
@@ -5942,6 +5922,9 @@ LLVM_Util::ptx_compile_group(llvm::Module* lib_module, const std::string& name,
     delete linked_module;
 
     return true;
+#else
+    return false;
+#endif
 }
 
 

src/testrender/testrender.cpp

@@ -42,6 +42,7 @@ static bool saveptx              = false;
 static bool warmup               = false;
 static bool profile              = false;
 static bool O0 = false, O1 = false, O2 = false;
+static int llvm_opt              = 1;  // LLVM optimization level
 static bool debugnan             = false;
 static bool debug_uninit         = false;
 static bool userdata_isconnected = false;
@@ -76,6 +77,15 @@ set_shadingsys_options()
     if (const char* opt_env = getenv("TESTSHADE_OPT"))  // overrides opt
         opt = atoi(opt_env);
     shadingsys->attribute("optimize", opt);
+
+    // The cost of more optimization passes usually pays for itself by
+    // reducing the number of instructions JIT ultimately has to lower to
+    // the target ISA.
+    if (const char* llvm_opt_env = getenv(
+            "TESTSHADE_LLVM_OPT"))  // overrides llvm_opt
+        llvm_opt = atoi(llvm_opt_env);
+    shadingsys->attribute("llvm_optimize", llvm_opt);
+
     shadingsys->attribute("profile", int(profile));
     shadingsys->attribute("lockgeom", 1);
     shadingsys->attribute("debug_nan", debugnan);
@@ -149,6 +159,8 @@ getargs(int argc, const char* argv[])
       .help("Do a little runtime shader optimization");
     ap.arg("-O2", &O2)
       .help("Do lots of runtime shader optimization");
+    ap.arg("--llvm_opt %d:LEVEL", &llvm_opt)
+      .help("LLVM JIT optimization level");
     ap.arg("--debugnan", &debugnan)
       .help("Turn on 'debugnan' mode");
     ap.arg("--path SEARCHPATH", &shaderpath)

testsuite/render-bumptest/run.py

@@ -5,4 +5,6 @@
 # https://github.com/AcademySoftwareFoundation/OpenShadingLanguage
 
 outputs = [ "out.exr" ]
-command = testrender("-r 256 256 -aa 4 bumptest.xml out.exr")
+command = testrender("-r 256 256 -aa 4 --llvm_opt 13 bumptest.xml out.exr")
+
+# Note: we pick this test arbitrarily as the one to verify llvm_opt=13 works

testsuite/render-cornell/run.py

@@ -7,4 +7,6 @@
 failthresh = 0.01
 failpercent = 1
 outputs = [ "out.exr" ]
-command = testrender("-r 256 256 -aa 4 cornell.xml out.exr")
+command = testrender("-r 256 256 -aa 4 --llvm_opt 12 cornell.xml out.exr")
+
+# Note: we pick this test arbitrarily as the one to verify llvm_opt=12 works