1; We run nvvm-reflect (and then optimize) this module twice, once with metadata 2; that enables FTZ, and again with metadata that disables it. 3 4; RUN: cat %s > %t.noftz 5; RUN: echo '!0 = !{i32 4, !"nvvm-reflect-ftz", i32 0}' >> %t.noftz 6; RUN: opt %t.noftz -S -mtriple=nvptx-nvidia-cuda -passes='nvvm-reflect,simplifycfg' \ 7; RUN: | FileCheck %s --check-prefix=USE_FTZ_0 --check-prefix=CHECK 8 9; RUN: cat %s > %t.ftz 10; RUN: echo '!0 = !{i32 4, !"nvvm-reflect-ftz", i32 1}' >> %t.ftz 11; RUN: opt %t.ftz -S -mtriple=nvptx-nvidia-cuda -passes='nvvm-reflect,simplifycfg' \ 12; RUN: | FileCheck %s --check-prefix=USE_FTZ_1 --check-prefix=CHECK 13 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: @foo 20define float @foo(float %a, float %b) { 21; CHECK-NOT: call i32 @__nvvm_reflect 22 %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, 0 25 br i1 %cmp, label %use_mul, label %use_add 26 27use_mul: 28; USE_FTZ_1: fmul float %a, %b 29; USE_FTZ_0-NOT: fadd float %a, %b 30 %ret1 = fmul float %a, %b 31 br label %exit 32 33use_add: 34; USE_FTZ_0: fadd float %a, %b 35; USE_FTZ_1-NOT: fmul float %a, %b 36 %ret2 = fadd float %a, %b 37 br label %exit 38 39exit: 40 %ret = phi float [%ret1, %use_mul], [%ret2, %use_add] 41 ret float %ret 42} 43 44declare i32 @llvm.nvvm.reflect(ptr) 45 46; CHECK-LABEL: define i32 @intrinsic 47 48define i32 @intrinsic() { 49; CHECK-NOT: call i32 @llvm.nvvm.reflect 50; USE_FTZ_0: ret i32 0 51; USE_FTZ_1: ret i32 1 52 %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 %reflect 55} 56 57; CUDA-7.0 passes __nvvm_reflect argument slightly differently. 58; Verify that it works, too 59 60@"$str" = private addrspace(1) constant [11 x i8] c"__CUDA_FTZ\00" 61 62; CHECK-LABEL: @bar 63define float @bar(float %a, float %b) { 64; CHECK-NOT: call i32 @__nvvm_reflect 65 %reflect = call i32 @__nvvm_reflect(ptr addrspacecast (ptr addrspace(1) @"$str" to ptr)) 66 %cmp = icmp ne i32 %reflect, 0 67 br i1 %cmp, label %use_mul, label %use_add 68 69use_mul: 70; USE_FTZ_1: fmul float %a, %b 71; USE_FTZ_0-NOT: fadd float %a, %b 72 %ret1 = fmul float %a, %b 73 br label %exit 74 75use_add: 76; USE_FTZ_0: fadd float %a, %b 77; USE_FTZ_1-NOT: fmul float %a, %b 78 %ret2 = fadd float %a, %b 79 br label %exit 80 81exit: 82 %ret = phi float [%ret1, %use_mul], [%ret2, %use_add] 83 ret float %ret 84} 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