88 lines · plain
1; We run nvvm-reflect (and then optimize) this module twice, once with metadata2; that enables FTZ, and again with metadata that disables it.3 4; RUN: cat %s > %t.noftz5; RUN: echo '!0 = !{i32 4, !"nvvm-reflect-ftz", i32 0}' >> %t.noftz6; RUN: opt %t.noftz -S -mtriple=nvptx-nvidia-cuda -passes='nvvm-reflect,simplifycfg' \7; RUN: | FileCheck %s --check-prefix=USE_FTZ_0 --check-prefix=CHECK8 9; RUN: cat %s > %t.ftz10; RUN: echo '!0 = !{i32 4, !"nvvm-reflect-ftz", i32 1}' >> %t.ftz11; RUN: opt %t.ftz -S -mtriple=nvptx-nvidia-cuda -passes='nvvm-reflect,simplifycfg' \12; RUN: | FileCheck %s --check-prefix=USE_FTZ_1 --check-prefix=CHECK13 14@str = private unnamed_addr addrspace(4) constant [11 x i8] c"__CUDA_FTZ\00"15 16declare i32 @__nvvm_reflect(ptr)17declare ptr @llvm.nvvm.ptr.constant.to.gen.p0.p4(ptr addrspace(4))18 19; CHECK-LABEL: @foo20define float @foo(float %a, float %b) {21; CHECK-NOT: call i32 @__nvvm_reflect22 %ptr = tail call ptr @llvm.nvvm.ptr.constant.to.gen.p0.p4(ptr addrspace(4) @str)23 %reflect = tail call i32 @__nvvm_reflect(ptr %ptr)24 %cmp = icmp ugt i32 %reflect, 025 br i1 %cmp, label %use_mul, label %use_add26 27use_mul:28; USE_FTZ_1: fmul float %a, %b29; USE_FTZ_0-NOT: fadd float %a, %b30 %ret1 = fmul float %a, %b31 br label %exit32 33use_add:34; USE_FTZ_0: fadd float %a, %b35; USE_FTZ_1-NOT: fmul float %a, %b36 %ret2 = fadd float %a, %b37 br label %exit38 39exit:40 %ret = phi float [%ret1, %use_mul], [%ret2, %use_add]41 ret float %ret42}43 44declare i32 @llvm.nvvm.reflect(ptr)45 46; CHECK-LABEL: define i32 @intrinsic47 48define i32 @intrinsic() {49; CHECK-NOT: call i32 @llvm.nvvm.reflect50; USE_FTZ_0: ret i32 051; USE_FTZ_1: ret i32 152 %ptr = tail call ptr @llvm.nvvm.ptr.constant.to.gen.p0.p4(ptr addrspace(4) @str)53 %reflect = tail call i32 @llvm.nvvm.reflect(ptr %ptr)54 ret i32 %reflect55}56 57; CUDA-7.0 passes __nvvm_reflect argument slightly differently.58; Verify that it works, too59 60@"$str" = private addrspace(1) constant [11 x i8] c"__CUDA_FTZ\00"61 62; CHECK-LABEL: @bar63define float @bar(float %a, float %b) {64; CHECK-NOT: call i32 @__nvvm_reflect65 %reflect = call i32 @__nvvm_reflect(ptr addrspacecast (ptr addrspace(1) @"$str" to ptr))66 %cmp = icmp ne i32 %reflect, 067 br i1 %cmp, label %use_mul, label %use_add68 69use_mul:70; USE_FTZ_1: fmul float %a, %b71; USE_FTZ_0-NOT: fadd float %a, %b72 %ret1 = fmul float %a, %b73 br label %exit74 75use_add:76; USE_FTZ_0: fadd float %a, %b77; USE_FTZ_1-NOT: fmul float %a, %b78 %ret2 = fadd float %a, %b79 br label %exit80 81exit:82 %ret = phi float [%ret1, %use_mul], [%ret2, %use_add]83 ret float %ret84}85 86!llvm.module.flags = !{!0}87; A module flag is added to the end of this file by the RUN lines at the top.88