[llvm] 0f070be - [NVPTX][FIX] Allow __nvvm_reflect in the presence of opaque pointers

[Johannes Doerfert via llvm-commits]

Author: Johannes Doerfert
Date: 2022-04-12T16:42:50-05:00
New Revision: 0f070bee8254678a3501ccb9a3f37933fe517ca7

URL: https://github.com/llvm/llvm-project/commit/0f070bee8254678a3501ccb9a3f37933fe517ca7
DIFF: https://github.com/llvm/llvm-project/commit/0f070bee8254678a3501ccb9a3f37933fe517ca7.diff

LOG: [NVPTX][FIX] Allow __nvvm_reflect in the presence of opaque pointers

Differential Revision: https://reviews.llvm.org/D123522

Added: 
    llvm/test/CodeGen/NVPTX/nvvm-reflect-opaque.ll

Modified: 
    llvm/lib/Target/NVPTX/NVVMReflect.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/NVPTX/NVVMReflect.cpp b/llvm/lib/Target/NVPTX/NVVMReflect.cpp
index 339f51d210874..3f3c4967609aa 100644
--- a/llvm/lib/Target/NVPTX/NVVMReflect.cpp
+++ b/llvm/lib/Target/NVPTX/NVVMReflect.cpp
@@ -133,15 +133,13 @@ static bool runNVVMReflect(Function &F, unsigned SmVersion) {
       // FIXME: Add assertions about ConvCall.
       Str = ConvCall->getArgOperand(0);
     }
-    assert(isa<ConstantExpr>(Str) &&
-           "Format of __nvvm__reflect function not recognized");
-    const ConstantExpr *GEP = cast<ConstantExpr>(Str);
-
-    const Value *Sym = GEP->getOperand(0);
-    assert(isa<Constant>(Sym) &&
+    // Pre opaque pointers we have a constant expression wrapping the constant
+    // string.
+    Str = Str->stripPointerCasts();
+    assert(isa<Constant>(Str) &&
            "Format of __nvvm_reflect function not recognized");
 
-    const Value *Operand = cast<Constant>(Sym)->getOperand(0);
+    const Value *Operand = cast<Constant>(Str)->getOperand(0);
     if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(Operand)) {
       // For CUDA-7.0 style __nvvm_reflect calls, we need to find the operand's
       // initializer.

diff  --git a/llvm/test/CodeGen/NVPTX/nvvm-reflect-opaque.ll b/llvm/test/CodeGen/NVPTX/nvvm-reflect-opaque.ll
new file mode 100644
index 0000000000000..73831603df941
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/nvvm-reflect-opaque.ll
@@ -0,0 +1,112 @@
+; We run nvvm-reflect (and then optimize) this module twice, once with metadata
+; that enables FTZ, and again with metadata that disables it.
+
+; RUN: cat %s > %t.noftz
+; RUN: echo '!0 = !{i32 4, !"nvvm-reflect-ftz", i32 0}' >> %t.noftz
+; RUN: opt %t.noftz -S -mtriple=nvptx-nvidia-cuda -passes='default<O2>' \
+; RUN:   | FileCheck %s --check-prefix=USE_FTZ_0 --check-prefix=CHECK
+
+; RUN: cat %s > %t.ftz
+; RUN: echo '!0 = !{i32 4, !"nvvm-reflect-ftz", i32 1}' >> %t.ftz
+; RUN: opt %t.ftz -S -mtriple=nvptx-nvidia-cuda -passes='default<O2>' \
+; RUN:   | FileCheck %s --check-prefix=USE_FTZ_1 --check-prefix=CHECK
+
+ at str = private unnamed_addr addrspace(4) constant [11 x i8] c"__CUDA_FTZ\00"
+
+declare i32 @__nvvm_reflect(ptr)
+declare ptr @llvm.nvvm.ptr.constant.to.gen.p0i8.p4i8(ptr addrspace(4))
+
+; CHECK-LABEL: @foo
+define float @foo(float %a, float %b) {
+; CHECK-NOT: call i32 @__nvvm_reflect
+  %ptr = tail call ptr @llvm.nvvm.ptr.constant.to.gen.p0i8.p4i8(ptr addrspace(4) @str)
+  %reflect = tail call i32 @__nvvm_reflect(ptr %ptr)
+  %cmp = icmp ugt i32 %reflect, 0
+  br i1 %cmp, label %use_mul, label %use_add
+
+use_mul:
+; USE_FTZ_1: fmul float %a, %b
+; USE_FTZ_0-NOT: fadd float %a, %b
+  %ret1 = fmul float %a, %b
+  br label %exit
+
+use_add:
+; USE_FTZ_0: fadd float %a, %b
+; USE_FTZ_1-NOT: fmul float %a, %b
+  %ret2 = fadd float %a, %b
+  br label %exit
+
+exit:
+  %ret = phi float [%ret1, %use_mul], [%ret2, %use_add]
+  ret float %ret
+}
+
+declare i32 @llvm.nvvm.reflect.p0i8(ptr)
+
+; CHECK-LABEL: define i32 @intrinsic
+define i32 @intrinsic() {
+; CHECK-NOT: call i32 @llvm.nvvm.reflect
+; USE_FTZ_0: ret i32 0
+; USE_FTZ_1: ret i32 1
+  %ptr = tail call ptr @llvm.nvvm.ptr.constant.to.gen.p0i8.p4i8(ptr addrspace(4) @str)
+  %reflect = tail call i32 @llvm.nvvm.reflect.p0i8(ptr %ptr)
+  ret i32 %reflect
+}
+
+; CUDA-7.0 passes __nvvm_reflect argument slightly 
diff erently.
+; Verify that it works, too
+
+@"$str" = private addrspace(1) constant [11 x i8] c"__CUDA_FTZ\00"
+
+; CHECK-LABEL: @bar
+define float @bar(float %a, float %b) {
+; CHECK-NOT: call i32 @__nvvm_reflect
+  %reflect = call i32 @__nvvm_reflect(ptr addrspacecast (ptr addrspace(1) @"$str" to ptr))
+  %cmp = icmp ne i32 %reflect, 0
+  br i1 %cmp, label %use_mul, label %use_add
+
+use_mul:
+; USE_FTZ_1: fmul float %a, %b
+; USE_FTZ_0-NOT: fadd float %a, %b
+  %ret1 = fmul float %a, %b
+  br label %exit
+
+use_add:
+; USE_FTZ_0: fadd float %a, %b
+; USE_FTZ_1-NOT: fmul float %a, %b
+  %ret2 = fadd float %a, %b
+  br label %exit
+
+exit:
+  %ret = phi float [%ret1, %use_mul], [%ret2, %use_add]
+  ret float %ret
+}
+
+ at str0 = private constant [11 x i8] c"__CUDA_FTZ\00"
+
+; CHECK-LABEL: @baz
+define float @baz(float %a, float %b) {
+; CHECK-NOT: call i32 @__nvvm_reflect
+  %reflect = call i32 @__nvvm_reflect(ptr @str0)
+  %cmp = icmp ne i32 %reflect, 0
+  br i1 %cmp, label %use_mul, label %use_add
+
+use_mul:
+; USE_FTZ_1: fmul float %a, %b
+; USE_FTZ_0-NOT: fadd float %a, %b
+  %ret1 = fmul float %a, %b
+  br label %exit
+
+use_add:
+; USE_FTZ_0: fadd float %a, %b
+; USE_FTZ_1-NOT: fmul float %a, %b
+  %ret2 = fadd float %a, %b
+  br label %exit
+
+exit:
+  %ret = phi float [%ret1, %use_mul], [%ret2, %use_add]
+  ret float %ret
+}
+
+!llvm.module.flags = !{!0}
+; A module flag is added to the end of this file by the RUN lines at the top.