Prevent SIMD size change with StackOverflowDetection

Do not count stackoverflow detection methods when checking if kernel has subroutine.

Author: ficol

IGC/Compiler/CISACodeGen/GenCodeGenModule.h

@@ -143,6 +143,14 @@ namespace IGC {
         bool isSingle() const {
             return (Functions.size() == 1 && Functions.front()->size() == 1);
         }
+        /// \brief Only one function in this group ignoring stack overflow detection methods
+        bool isSingleIgnoringStackOverflowDetection() const {
+            auto isNotStackOverflowDetection = [](const llvm::Function* F) {
+                return !F->getName().startswith("__stackoverflow_detection") && !F->getName().startswith("__stackoverflow_init");
+            };
+            return (Functions.size() == 1 &&
+                    std::count_if(Functions.front()->begin(), Functions.front()->end(), isNotStackOverflowDetection) == 1);
+        }
         /// \brief Function group has a subroutine
         bool hasSubroutine() const {
             return m_hasSubroutine;

IGC/Compiler/CISACodeGen/OpenCLKernelCodeGen.cpp

@@ -3700,7 +3700,7 @@ namespace IGC
         auto FG = m_FGA ? m_FGA->getGroup(&F) : nullptr;
         bool hasStackCall = FG && FG->hasStackCall();
         bool isIndirectGroup = FG && m_FGA->isIndirectCallGroup(FG);
-        bool hasSubroutine = FG && !FG->isSingle() && !hasStackCall && !isIndirectGroup;
+        bool hasSubroutine = FG && !FG->isSingleIgnoringStackOverflowDetection() && !hasStackCall && !isIndirectGroup;
         bool forceLowestSIMDForStackCalls = IGC_IS_FLAG_ENABLED(ForceLowestSIMDForStackCalls) && (hasStackCall || isIndirectGroup);
 
 

IGC/ocloc_tests/features/stack_overflow_detection/stack_overflow_detection_no_vla.ll

@@ -29,8 +29,8 @@
 ; RUN: ocloc compile -spirv_input -file %t.spv -options " -igc_opts 'DumpVISAASMToConsole=1,StackOverflowDetection=1'" -device pvc | FileCheck %s --check-prefix=CHECK-VISA
 
 ; CHECK-VISA: .kernel "test_simple"
-; CHECK-VISA-NOT: call (M1, {{(8)|(16)}}) __stackoverflow_init{{.*}}
-; CHECK-VISA-NOT: call (M1, {{(8)|(16)}}) __stackoverflow_detection{{.*}}
+; CHECK-VISA-NOT: call (M1, {{(8)|(16)|(32)}}) __stackoverflow_init{{.*}}
+; CHECK-VISA-NOT: call (M1, {{(8)|(16)|(32)}}) __stackoverflow_detection{{.*}}
 
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"

IGC/ocloc_tests/features/stack_overflow_detection/stack_overflow_detection_vla.ll

@@ -19,9 +19,9 @@
 ; RUN: ocloc compile -spirv_input -file %t.spv -options " -igc_opts 'DumpVISAASMToConsole=1,StackOverflowDetection=1'" -device pvc | FileCheck %s --check-prefix=CHECK-VISA
 
 ; CHECK-VISA: .kernel "test_simple"
-; CHECK-VISA: call (M1, {{(8)|(16)}}) __stackoverflow_init{{.*}}
-; CHECK-VISA-NOT: call (M1, {{(8)|(16)}}) __stackoverflow_init{{.*}}
-; CHECK-VISA: call (M1, {{(8)|(16)}}) __stackoverflow_detection{{.*}}
+; CHECK-VISA: call (M1, {{(8)|(16)|(32)}}) __stackoverflow_init{{.*}}
+; CHECK-VISA-NOT: call (M1, {{(8)|(16)|(32)}}) __stackoverflow_init{{.*}}
+; CHECK-VISA: call (M1, {{(8)|(16)|(32)}}) __stackoverflow_detection{{.*}}
 ;
 ; CHECK-VISA: warning: VLA has been detected
 

IGC/ocloc_tests/features/stack_overflow_detection/stack_overflow_simdsize.cl

@@ -0,0 +1,32 @@
+/*========================== begin_copyright_notice ============================
+
+Copyright (C) 2024 Intel Corporation
+
+SPDX-License-Identifier: MIT
+
+============================= end_copyright_notice ===========================*/
+// This test checks if calls to stack overflow detection subroutines don't change kernel SIMD size.
+
+// REQUIRES: regkeys, pvc-supported
+
+// RUN: ocloc compile -file %s -options " -igc_opts 'DumpVISAASMToConsole=1,StackOverflowDetection=1'" -device pvc | FileCheck %s --check-prefix=CHECK-VISA
+
+// CHECK-VISA: .kernel "test_nosubroutines"
+// CHECK-VISA-NOT: .function
+// CHECK-VISA: .kernel_attr SimdSize=32
+
+// CHECK-VISA: .kernel "test_subroutines"
+// CHECK-VISA-NOT: .function
+// CHECK-VISA: .kernel_attr SimdSize=16
+
+int fact(int n) {
+  return n < 2 ? 1 : n*fact(n-1);
+}
+
+kernel void test_subroutines(global int* out, int n) {
+  out[0] = fact(n);
+}
+
+kernel void test_nosubroutines(global int* out, int n) {
+  out[0] = n;
+}