diff options
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll | 194 |
1 files changed, 97 insertions, 97 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll index d9e0464..6bda962 100644 --- a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll @@ -12,7 +12,7 @@ define amdgpu_kernel void @noop_fdiv_fpmath(ptr addrspace(1) %out, float %a, float %b) #0 { ; CHECK-LABEL: define amdgpu_kernel void @noop_fdiv_fpmath( ; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]]) #[[ATTR0:[0-9]+]] { -; CHECK-NEXT: [[MD_25ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath !0 +; CHECK-NEXT: [[MD_25ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META0:![0-9]+]] ; CHECK-NEXT: store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; CHECK-NEXT: ret void ; @@ -26,7 +26,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32(ptr addrspace(1) %out, float %a, floa ; IEEE-GOODFREXP-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]]) #[[ATTR1:[0-9]+]] { ; IEEE-GOODFREXP-NEXT: [[NO_MD:%.*]] = fdiv float [[A]], [[B]] ; IEEE-GOODFREXP-NEXT: store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath !1 +; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1:![0-9]+]] ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[B]]) ; IEEE-GOODFREXP-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 @@ -61,9 +61,9 @@ define amdgpu_kernel void @fdiv_fpmath_f32(ptr addrspace(1) %out, float %a, floa ; IEEE-GOODFREXP-NEXT: [[TMP27:%.*]] = sub i32 [[TMP25]], [[TMP21]] ; IEEE-GOODFREXP-NEXT: [[MD_3ULP:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP26]], i32 [[TMP27]]) ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath !0 +; IEEE-GOODFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath [[META0]] ; IEEE-GOODFREXP-NEXT: store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath !0 +; IEEE-GOODFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath [[META0]] ; IEEE-GOODFREXP-NEXT: store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[NO_MD_ARCP:%.*]] = fdiv arcp float [[A]], [[B]] ; IEEE-GOODFREXP-NEXT: store volatile float [[NO_MD_ARCP]], ptr addrspace(1) [[OUT]], align 4 @@ -89,7 +89,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32(ptr addrspace(1) %out, float %a, floa ; IEEE-BADFREXP-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]]) #[[ATTR1:[0-9]+]] { ; IEEE-BADFREXP-NEXT: [[NO_MD:%.*]] = fdiv float [[A]], [[B]] ; IEEE-BADFREXP-NEXT: store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath !1 +; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1:![0-9]+]] ; IEEE-BADFREXP-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[B]]) ; IEEE-BADFREXP-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 @@ -124,9 +124,9 @@ define amdgpu_kernel void @fdiv_fpmath_f32(ptr addrspace(1) %out, float %a, floa ; IEEE-BADFREXP-NEXT: [[TMP27:%.*]] = sub i32 [[TMP25]], [[TMP21]] ; IEEE-BADFREXP-NEXT: [[MD_3ULP:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP26]], i32 [[TMP27]]) ; IEEE-BADFREXP-NEXT: store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath !0 +; IEEE-BADFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath [[META0]] ; IEEE-BADFREXP-NEXT: store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath !0 +; IEEE-BADFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath [[META0]] ; IEEE-BADFREXP-NEXT: store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[NO_MD_ARCP:%.*]] = fdiv arcp float [[A]], [[B]] ; IEEE-BADFREXP-NEXT: store volatile float [[NO_MD_ARCP]], ptr addrspace(1) [[OUT]], align 4 @@ -152,7 +152,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32(ptr addrspace(1) %out, float %a, floa ; DAZ-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]]) #[[ATTR1:[0-9]+]] { ; DAZ-NEXT: [[NO_MD:%.*]] = fdiv float [[A]], [[B]] ; DAZ-NEXT: store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath !1 +; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1:![0-9]+]] ; DAZ-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[B]]) ; DAZ-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 @@ -169,9 +169,9 @@ define amdgpu_kernel void @fdiv_fpmath_f32(ptr addrspace(1) %out, float %a, floa ; DAZ-NEXT: store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[MD_3ULP:%.*]] = call float @llvm.amdgcn.fdiv.fast(float [[A]], float [[B]]) ; DAZ-NEXT: store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath !0 +; DAZ-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath [[META0]] ; DAZ-NEXT: store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath !0 +; DAZ-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath [[META0]] ; DAZ-NEXT: store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[NO_MD_ARCP:%.*]] = fdiv arcp float [[A]], [[B]] ; DAZ-NEXT: store volatile float [[NO_MD_ARCP]], ptr addrspace(1) [[OUT]], align 4 @@ -430,15 +430,15 @@ define amdgpu_kernel void @rcp_fdiv_f32_fpmath(ptr addrspace(1) %out, float %x) ; IEEE-GOODFREXP-NEXT: [[TMP10:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[TMP7]]) ; IEEE-GOODFREXP-NEXT: [[MD_25ULP:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP10]], i32 [[TMP9]]) ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float 1.000000e+00, [[X]], !fpmath !1 +; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float 1.000000e+00, [[X]], !fpmath [[META1]] ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[AFN_NO_MD:%.*]] = fdiv afn float 1.000000e+00, [[X]] ; IEEE-GOODFREXP-NEXT: store volatile float [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[AFN_25ULP:%.*]] = fdiv afn float 1.000000e+00, [[X]], !fpmath !0 +; IEEE-GOODFREXP-NEXT: [[AFN_25ULP:%.*]] = fdiv afn float 1.000000e+00, [[X]], !fpmath [[META0]] ; IEEE-GOODFREXP-NEXT: store volatile float [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[FAST_NO_MD:%.*]] = fdiv fast float 1.000000e+00, [[X]] ; IEEE-GOODFREXP-NEXT: store volatile float [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[FAST_25ULP:%.*]] = fdiv fast float 1.000000e+00, [[X]], !fpmath !0 +; IEEE-GOODFREXP-NEXT: [[FAST_25ULP:%.*]] = fdiv fast float 1.000000e+00, [[X]], !fpmath [[META0]] ; IEEE-GOODFREXP-NEXT: store volatile float [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[TMP11:%.*]] = fneg float [[X]] ; IEEE-GOODFREXP-NEXT: [[TMP12:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP11]]) @@ -458,7 +458,7 @@ define amdgpu_kernel void @rcp_fdiv_f32_fpmath(ptr addrspace(1) %out, float %x) ; IEEE-GOODFREXP-NEXT: store volatile float [[NEG_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[NEG_AFN_NO_MD:%.*]] = fdiv afn float -1.000000e+00, [[X]] ; IEEE-GOODFREXP-NEXT: store volatile float [[NEG_AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[NEG_AFN_25ULP:%.*]] = fdiv afn float -1.000000e+00, [[X]], !fpmath !0 +; IEEE-GOODFREXP-NEXT: [[NEG_AFN_25ULP:%.*]] = fdiv afn float -1.000000e+00, [[X]], !fpmath [[META0]] ; IEEE-GOODFREXP-NEXT: store volatile float [[NEG_AFN_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[NEG_FAST_NO_MD:%.*]] = fdiv fast float -1.000000e+00, [[X]] ; IEEE-GOODFREXP-NEXT: store volatile float [[NEG_FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 4 @@ -482,15 +482,15 @@ define amdgpu_kernel void @rcp_fdiv_f32_fpmath(ptr addrspace(1) %out, float %x) ; IEEE-BADFREXP-NEXT: [[TMP10:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[TMP7]]) ; IEEE-BADFREXP-NEXT: [[MD_25ULP:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP10]], i32 [[TMP9]]) ; IEEE-BADFREXP-NEXT: store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float 1.000000e+00, [[X]], !fpmath !1 +; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float 1.000000e+00, [[X]], !fpmath [[META1]] ; IEEE-BADFREXP-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[AFN_NO_MD:%.*]] = fdiv afn float 1.000000e+00, [[X]] ; IEEE-BADFREXP-NEXT: store volatile float [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[AFN_25ULP:%.*]] = fdiv afn float 1.000000e+00, [[X]], !fpmath !0 +; IEEE-BADFREXP-NEXT: [[AFN_25ULP:%.*]] = fdiv afn float 1.000000e+00, [[X]], !fpmath [[META0]] ; IEEE-BADFREXP-NEXT: store volatile float [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[FAST_NO_MD:%.*]] = fdiv fast float 1.000000e+00, [[X]] ; IEEE-BADFREXP-NEXT: store volatile float [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[FAST_25ULP:%.*]] = fdiv fast float 1.000000e+00, [[X]], !fpmath !0 +; IEEE-BADFREXP-NEXT: [[FAST_25ULP:%.*]] = fdiv fast float 1.000000e+00, [[X]], !fpmath [[META0]] ; IEEE-BADFREXP-NEXT: store volatile float [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[TMP11:%.*]] = fneg float [[X]] ; IEEE-BADFREXP-NEXT: [[TMP12:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP11]]) @@ -510,7 +510,7 @@ define amdgpu_kernel void @rcp_fdiv_f32_fpmath(ptr addrspace(1) %out, float %x) ; IEEE-BADFREXP-NEXT: store volatile float [[NEG_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[NEG_AFN_NO_MD:%.*]] = fdiv afn float -1.000000e+00, [[X]] ; IEEE-BADFREXP-NEXT: store volatile float [[NEG_AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[NEG_AFN_25ULP:%.*]] = fdiv afn float -1.000000e+00, [[X]], !fpmath !0 +; IEEE-BADFREXP-NEXT: [[NEG_AFN_25ULP:%.*]] = fdiv afn float -1.000000e+00, [[X]], !fpmath [[META0]] ; IEEE-BADFREXP-NEXT: store volatile float [[NEG_AFN_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[NEG_FAST_NO_MD:%.*]] = fdiv fast float -1.000000e+00, [[X]] ; IEEE-BADFREXP-NEXT: store volatile float [[NEG_FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 4 @@ -524,15 +524,15 @@ define amdgpu_kernel void @rcp_fdiv_f32_fpmath(ptr addrspace(1) %out, float %x) ; DAZ-NEXT: store volatile float [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[MD_25ULP:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[X]]) ; DAZ-NEXT: store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float 1.000000e+00, [[X]], !fpmath !1 +; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float 1.000000e+00, [[X]], !fpmath [[META1]] ; DAZ-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[AFN_NO_MD:%.*]] = fdiv afn float 1.000000e+00, [[X]] ; DAZ-NEXT: store volatile float [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[AFN_25ULP:%.*]] = fdiv afn float 1.000000e+00, [[X]], !fpmath !0 +; DAZ-NEXT: [[AFN_25ULP:%.*]] = fdiv afn float 1.000000e+00, [[X]], !fpmath [[META0]] ; DAZ-NEXT: store volatile float [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[FAST_NO_MD:%.*]] = fdiv fast float 1.000000e+00, [[X]] ; DAZ-NEXT: store volatile float [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[FAST_25ULP:%.*]] = fdiv fast float 1.000000e+00, [[X]], !fpmath !0 +; DAZ-NEXT: [[FAST_25ULP:%.*]] = fdiv fast float 1.000000e+00, [[X]], !fpmath [[META0]] ; DAZ-NEXT: store volatile float [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[TMP1:%.*]] = fneg float [[X]] ; DAZ-NEXT: [[NEG_MD_1ULP:%.*]] = call float @llvm.amdgcn.rcp.f32(float [[TMP1]]) @@ -542,7 +542,7 @@ define amdgpu_kernel void @rcp_fdiv_f32_fpmath(ptr addrspace(1) %out, float %x) ; DAZ-NEXT: store volatile float [[NEG_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[NEG_AFN_NO_MD:%.*]] = fdiv afn float -1.000000e+00, [[X]] ; DAZ-NEXT: store volatile float [[NEG_AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[NEG_AFN_25ULP:%.*]] = fdiv afn float -1.000000e+00, [[X]], !fpmath !0 +; DAZ-NEXT: [[NEG_AFN_25ULP:%.*]] = fdiv afn float -1.000000e+00, [[X]], !fpmath [[META0]] ; DAZ-NEXT: store volatile float [[NEG_AFN_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[NEG_FAST_NO_MD:%.*]] = fdiv fast float -1.000000e+00, [[X]] ; DAZ-NEXT: store volatile float [[NEG_FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 4 @@ -1179,7 +1179,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32_vector(ptr addrspace(1) %out, <2 x fl ; IEEE-GOODFREXP-SAME: ptr addrspace(1) [[OUT:%.*]], <2 x float> [[A:%.*]], <2 x float> [[B:%.*]]) #[[ATTR1]] { ; IEEE-GOODFREXP-NEXT: [[NO_MD:%.*]] = fdiv <2 x float> [[A]], [[B]] ; IEEE-GOODFREXP-NEXT: store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 8 -; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv <2 x float> [[A]], [[B]], !fpmath !1 +; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv <2 x float> [[A]], [[B]], !fpmath [[META1]] ; IEEE-GOODFREXP-NEXT: store volatile <2 x float> [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 8 ; IEEE-GOODFREXP-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[A]], i64 0 ; IEEE-GOODFREXP-NEXT: [[TMP2:%.*]] = extractelement <2 x float> [[A]], i64 1 @@ -1241,7 +1241,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32_vector(ptr addrspace(1) %out, <2 x fl ; IEEE-BADFREXP-SAME: ptr addrspace(1) [[OUT:%.*]], <2 x float> [[A:%.*]], <2 x float> [[B:%.*]]) #[[ATTR1]] { ; IEEE-BADFREXP-NEXT: [[NO_MD:%.*]] = fdiv <2 x float> [[A]], [[B]] ; IEEE-BADFREXP-NEXT: store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 8 -; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv <2 x float> [[A]], [[B]], !fpmath !1 +; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv <2 x float> [[A]], [[B]], !fpmath [[META1]] ; IEEE-BADFREXP-NEXT: store volatile <2 x float> [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 8 ; IEEE-BADFREXP-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[A]], i64 0 ; IEEE-BADFREXP-NEXT: [[TMP2:%.*]] = extractelement <2 x float> [[A]], i64 1 @@ -1303,7 +1303,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32_vector(ptr addrspace(1) %out, <2 x fl ; DAZ-SAME: ptr addrspace(1) [[OUT:%.*]], <2 x float> [[A:%.*]], <2 x float> [[B:%.*]]) #[[ATTR1]] { ; DAZ-NEXT: [[NO_MD:%.*]] = fdiv <2 x float> [[A]], [[B]] ; DAZ-NEXT: store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 8 -; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv <2 x float> [[A]], [[B]], !fpmath !1 +; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv <2 x float> [[A]], [[B]], !fpmath [[META1]] ; DAZ-NEXT: store volatile <2 x float> [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 8 ; DAZ-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[A]], i64 0 ; DAZ-NEXT: [[TMP2:%.*]] = extractelement <2 x float> [[A]], i64 1 @@ -1359,15 +1359,15 @@ define amdgpu_kernel void @rcp_fdiv_f32_vector_fpmath(ptr addrspace(1) %out, <2 ; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], <2 x float> [[X:%.*]]) #[[ATTR1:[0-9]+]] { ; CHECK-NEXT: [[NO_MD:%.*]] = fdiv <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[X]] ; CHECK-NEXT: store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 8 -; CHECK-NEXT: [[MD_HALF_ULP:%.*]] = fdiv <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[X]], !fpmath !1 +; CHECK-NEXT: [[MD_HALF_ULP:%.*]] = fdiv <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[X]], !fpmath [[META1:![0-9]+]] ; CHECK-NEXT: store volatile <2 x float> [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 8 ; CHECK-NEXT: [[AFN_NO_MD:%.*]] = fdiv afn <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[X]] ; CHECK-NEXT: store volatile <2 x float> [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 8 ; CHECK-NEXT: [[FAST_NO_MD:%.*]] = fdiv fast <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[X]] ; CHECK-NEXT: store volatile <2 x float> [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 8 -; CHECK-NEXT: [[AFN_25ULP:%.*]] = fdiv afn <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[X]], !fpmath !0 +; CHECK-NEXT: [[AFN_25ULP:%.*]] = fdiv afn <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[X]], !fpmath [[META0]] ; CHECK-NEXT: store volatile <2 x float> [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 8 -; CHECK-NEXT: [[FAST_25ULP:%.*]] = fdiv fast <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[X]], !fpmath !0 +; CHECK-NEXT: [[FAST_25ULP:%.*]] = fdiv fast <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[X]], !fpmath [[META0]] ; CHECK-NEXT: store volatile <2 x float> [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 8 ; CHECK-NEXT: ret void ; @@ -1395,9 +1395,9 @@ define amdgpu_kernel void @rcp_fdiv_f32_fpmath_vector_nonsplat(ptr addrspace(1) ; CHECK-NEXT: store volatile <2 x float> [[AFN_NO_MD]], ptr addrspace(1) [[OUT]], align 8 ; CHECK-NEXT: [[FAST_NO_MD:%.*]] = fdiv fast <2 x float> <float 1.000000e+00, float 2.000000e+00>, [[X]] ; CHECK-NEXT: store volatile <2 x float> [[FAST_NO_MD]], ptr addrspace(1) [[OUT]], align 8 -; CHECK-NEXT: [[AFN_25ULP:%.*]] = fdiv afn <2 x float> <float 1.000000e+00, float 2.000000e+00>, [[X]], !fpmath !0 +; CHECK-NEXT: [[AFN_25ULP:%.*]] = fdiv afn <2 x float> <float 1.000000e+00, float 2.000000e+00>, [[X]], !fpmath [[META0]] ; CHECK-NEXT: store volatile <2 x float> [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 8 -; CHECK-NEXT: [[FAST_25ULP:%.*]] = fdiv fast <2 x float> <float 1.000000e+00, float 2.000000e+00>, [[X]], !fpmath !0 +; CHECK-NEXT: [[FAST_25ULP:%.*]] = fdiv fast <2 x float> <float 1.000000e+00, float 2.000000e+00>, [[X]], !fpmath [[META0]] ; CHECK-NEXT: store volatile <2 x float> [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 8 ; CHECK-NEXT: ret void ; @@ -1418,9 +1418,9 @@ define amdgpu_kernel void @rcp_fdiv_f32_vector_fpmath_partial_constant(ptr addrs ; CHECK-LABEL: define amdgpu_kernel void @rcp_fdiv_f32_vector_fpmath_partial_constant( ; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], <2 x float> [[X:%.*]], <2 x float> [[Y:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[X_INSERT:%.*]] = insertelement <2 x float> [[X]], float 1.000000e+00, i32 0 -; CHECK-NEXT: [[AFN_25ULP:%.*]] = fdiv afn <2 x float> [[X_INSERT]], [[Y]], !fpmath !0 +; CHECK-NEXT: [[AFN_25ULP:%.*]] = fdiv afn <2 x float> [[X_INSERT]], [[Y]], !fpmath [[META0]] ; CHECK-NEXT: store volatile <2 x float> [[AFN_25ULP]], ptr addrspace(1) [[OUT]], align 8 -; CHECK-NEXT: [[FAST_25ULP:%.*]] = fdiv fast <2 x float> [[X_INSERT]], [[Y]], !fpmath !0 +; CHECK-NEXT: [[FAST_25ULP:%.*]] = fdiv fast <2 x float> [[X_INSERT]], [[Y]], !fpmath [[META0]] ; CHECK-NEXT: store volatile <2 x float> [[FAST_25ULP]], ptr addrspace(1) [[OUT]], align 8 ; CHECK-NEXT: ret void ; @@ -1520,7 +1520,7 @@ define amdgpu_kernel void @rsq_f32_fpmath(ptr addrspace(1) %out, float %x) { ; IEEE-GOODFREXP-NEXT: [[TMP5:%.*]] = select contract i1 [[TMP1]], float 4.096000e+03, float 1.000000e+00 ; IEEE-GOODFREXP-NEXT: [[MD_1ULP:%.*]] = fmul contract float [[TMP4]], [[TMP5]] ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[SQRT_MD_1ULP_MULTI_USE:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath !2 +; IEEE-GOODFREXP-NEXT: [[SQRT_MD_1ULP_MULTI_USE:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath [[META2:![0-9]+]] ; IEEE-GOODFREXP-NEXT: store volatile float [[SQRT_MD_1ULP_MULTI_USE]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[TMP6:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[SQRT_MD_1ULP_MULTI_USE]]) ; IEEE-GOODFREXP-NEXT: [[TMP7:%.*]] = extractvalue { float, i32 } [[TMP6]], 0 @@ -1536,8 +1536,8 @@ define amdgpu_kernel void @rsq_f32_fpmath(ptr addrspace(1) %out, float %x) { ; IEEE-GOODFREXP-NEXT: [[TMP15:%.*]] = select contract i1 [[TMP11]], float 4.096000e+03, float 1.000000e+00 ; IEEE-GOODFREXP-NEXT: [[MD_25ULP:%.*]] = fmul contract float [[TMP14]], [[TMP15]] ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[SQRT_MD_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath !1 -; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv contract float 1.000000e+00, [[SQRT_MD_HALF_ULP]], !fpmath !1 +; IEEE-GOODFREXP-NEXT: [[SQRT_MD_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath [[META1]] +; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv contract float 1.000000e+00, [[SQRT_MD_HALF_ULP]], !fpmath [[META1]] ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[SQRT_X_AFN_NO_MD:%.*]] = call contract afn float @llvm.sqrt.f32(float [[X]]) ; IEEE-GOODFREXP-NEXT: [[AFN_NO_MD:%.*]] = fdiv contract afn float 1.000000e+00, [[SQRT_X_AFN_NO_MD]] @@ -1563,7 +1563,7 @@ define amdgpu_kernel void @rsq_f32_fpmath(ptr addrspace(1) %out, float %x) { ; IEEE-GOODFREXP-NEXT: [[TMP25:%.*]] = select contract i1 [[TMP21]], float -4.096000e+03, float -1.000000e+00 ; IEEE-GOODFREXP-NEXT: [[NEG_FDIV_OPENCL:%.*]] = fmul contract float [[TMP24]], [[TMP25]] ; IEEE-GOODFREXP-NEXT: store volatile float [[NEG_FDIV_OPENCL]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[SQRT_X_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath !1 +; IEEE-GOODFREXP-NEXT: [[SQRT_X_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath [[META1]] ; IEEE-GOODFREXP-NEXT: [[TMP26:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[SQRT_X_HALF_ULP]]) ; IEEE-GOODFREXP-NEXT: [[TMP27:%.*]] = extractvalue { float, i32 } [[TMP26]], 0 ; IEEE-GOODFREXP-NEXT: [[TMP28:%.*]] = extractvalue { float, i32 } [[TMP26]], 1 @@ -1601,7 +1601,7 @@ define amdgpu_kernel void @rsq_f32_fpmath(ptr addrspace(1) %out, float %x) { ; IEEE-BADFREXP-NEXT: [[TMP5:%.*]] = select contract i1 [[TMP1]], float 4.096000e+03, float 1.000000e+00 ; IEEE-BADFREXP-NEXT: [[MD_1ULP:%.*]] = fmul contract float [[TMP4]], [[TMP5]] ; IEEE-BADFREXP-NEXT: store volatile float [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[SQRT_MD_1ULP_MULTI_USE:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath !2 +; IEEE-BADFREXP-NEXT: [[SQRT_MD_1ULP_MULTI_USE:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath [[META2:![0-9]+]] ; IEEE-BADFREXP-NEXT: store volatile float [[SQRT_MD_1ULP_MULTI_USE]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[TMP6:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[SQRT_MD_1ULP_MULTI_USE]]) ; IEEE-BADFREXP-NEXT: [[TMP7:%.*]] = extractvalue { float, i32 } [[TMP6]], 0 @@ -1617,8 +1617,8 @@ define amdgpu_kernel void @rsq_f32_fpmath(ptr addrspace(1) %out, float %x) { ; IEEE-BADFREXP-NEXT: [[TMP15:%.*]] = select contract i1 [[TMP11]], float 4.096000e+03, float 1.000000e+00 ; IEEE-BADFREXP-NEXT: [[MD_25ULP:%.*]] = fmul contract float [[TMP14]], [[TMP15]] ; IEEE-BADFREXP-NEXT: store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[SQRT_MD_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath !1 -; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv contract float 1.000000e+00, [[SQRT_MD_HALF_ULP]], !fpmath !1 +; IEEE-BADFREXP-NEXT: [[SQRT_MD_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath [[META1]] +; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv contract float 1.000000e+00, [[SQRT_MD_HALF_ULP]], !fpmath [[META1]] ; IEEE-BADFREXP-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[SQRT_X_AFN_NO_MD:%.*]] = call contract afn float @llvm.sqrt.f32(float [[X]]) ; IEEE-BADFREXP-NEXT: [[AFN_NO_MD:%.*]] = fdiv contract afn float 1.000000e+00, [[SQRT_X_AFN_NO_MD]] @@ -1644,7 +1644,7 @@ define amdgpu_kernel void @rsq_f32_fpmath(ptr addrspace(1) %out, float %x) { ; IEEE-BADFREXP-NEXT: [[TMP25:%.*]] = select contract i1 [[TMP21]], float -4.096000e+03, float -1.000000e+00 ; IEEE-BADFREXP-NEXT: [[NEG_FDIV_OPENCL:%.*]] = fmul contract float [[TMP24]], [[TMP25]] ; IEEE-BADFREXP-NEXT: store volatile float [[NEG_FDIV_OPENCL]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[SQRT_X_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath !1 +; IEEE-BADFREXP-NEXT: [[SQRT_X_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath [[META1]] ; IEEE-BADFREXP-NEXT: [[TMP26:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[SQRT_X_HALF_ULP]]) ; IEEE-BADFREXP-NEXT: [[TMP27:%.*]] = extractvalue { float, i32 } [[TMP26]], 0 ; IEEE-BADFREXP-NEXT: [[TMP28:%.*]] = call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[SQRT_X_HALF_ULP]]) @@ -1683,8 +1683,8 @@ define amdgpu_kernel void @rsq_f32_fpmath(ptr addrspace(1) %out, float %x) { ; DAZ-NEXT: store volatile float [[MD_1ULP_MULTI_USE]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[MD_25ULP:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float [[X]]) ; DAZ-NEXT: store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[SQRT_MD_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath !1 -; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv contract float 1.000000e+00, [[SQRT_MD_HALF_ULP]], !fpmath !1 +; DAZ-NEXT: [[SQRT_MD_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath [[META1]] +; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv contract float 1.000000e+00, [[SQRT_MD_HALF_ULP]], !fpmath [[META1]] ; DAZ-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[SQRT_X_AFN_NO_MD:%.*]] = call contract afn float @llvm.sqrt.f32(float [[X]]) ; DAZ-NEXT: [[AFN_NO_MD:%.*]] = fdiv contract afn float 1.000000e+00, [[SQRT_X_AFN_NO_MD]] @@ -1701,7 +1701,7 @@ define amdgpu_kernel void @rsq_f32_fpmath(ptr addrspace(1) %out, float %x) { ; DAZ-NEXT: [[TMP1:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float [[X]]) ; DAZ-NEXT: [[NEG_FDIV_OPENCL:%.*]] = fneg contract float [[TMP1]] ; DAZ-NEXT: store volatile float [[NEG_FDIV_OPENCL]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[SQRT_X_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath !1 +; DAZ-NEXT: [[SQRT_X_HALF_ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath [[META1]] ; DAZ-NEXT: [[FDIV_SQRT_MISMATCH_MD0:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[SQRT_X_HALF_ULP]]) ; DAZ-NEXT: store volatile float [[FDIV_SQRT_MISMATCH_MD0]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[SQRT_MISMATCH_MD1:%.*]] = call afn float @llvm.sqrt.f32(float [[X]]) @@ -1873,7 +1873,7 @@ define amdgpu_kernel void @rsq_f32_fpmath_flags(ptr addrspace(1) %out, float %x) define float @rsq_f32_missing_contract0(float %x) { ; IEEE-GOODFREXP-LABEL: define float @rsq_f32_missing_contract0( ; IEEE-GOODFREXP-SAME: float [[X:%.*]]) #[[ATTR1]] { -; IEEE-GOODFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath !2 +; IEEE-GOODFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META2]] ; IEEE-GOODFREXP-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[SQRT_X_3ULP]]) ; IEEE-GOODFREXP-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 ; IEEE-GOODFREXP-NEXT: [[TMP3:%.*]] = extractvalue { float, i32 } [[TMP1]], 1 @@ -1884,7 +1884,7 @@ define float @rsq_f32_missing_contract0(float %x) { ; ; IEEE-BADFREXP-LABEL: define float @rsq_f32_missing_contract0( ; IEEE-BADFREXP-SAME: float [[X:%.*]]) #[[ATTR1]] { -; IEEE-BADFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath !2 +; IEEE-BADFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call float @llvm.sqrt.f32(float [[X]]), !fpmath [[META2]] ; IEEE-BADFREXP-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[SQRT_X_3ULP]]) ; IEEE-BADFREXP-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 ; IEEE-BADFREXP-NEXT: [[TMP3:%.*]] = call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[SQRT_X_3ULP]]) @@ -1907,7 +1907,7 @@ define float @rsq_f32_missing_contract0(float %x) { define float @rsq_f32_missing_contract1(float %x) { ; IEEE-GOODFREXP-LABEL: define float @rsq_f32_missing_contract1( ; IEEE-GOODFREXP-SAME: float [[X:%.*]]) #[[ATTR1]] { -; IEEE-GOODFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath !2 +; IEEE-GOODFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath [[META2]] ; IEEE-GOODFREXP-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[SQRT_X_3ULP]]) ; IEEE-GOODFREXP-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 ; IEEE-GOODFREXP-NEXT: [[TMP3:%.*]] = extractvalue { float, i32 } [[TMP1]], 1 @@ -1918,7 +1918,7 @@ define float @rsq_f32_missing_contract1(float %x) { ; ; IEEE-BADFREXP-LABEL: define float @rsq_f32_missing_contract1( ; IEEE-BADFREXP-SAME: float [[X:%.*]]) #[[ATTR1]] { -; IEEE-BADFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath !2 +; IEEE-BADFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call contract float @llvm.sqrt.f32(float [[X]]), !fpmath [[META2]] ; IEEE-BADFREXP-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[SQRT_X_3ULP]]) ; IEEE-BADFREXP-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 ; IEEE-BADFREXP-NEXT: [[TMP3:%.*]] = call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[SQRT_X_3ULP]]) @@ -2116,7 +2116,7 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; IEEE-GOODFREXP-NEXT: [[SQRT_X_NO_MD:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]) ; IEEE-GOODFREXP-NEXT: [[NO_MD:%.*]] = fdiv contract <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[SQRT_X_NO_MD]] ; IEEE-GOODFREXP-NEXT: store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[SQRT_MD_1ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath !2 +; IEEE-GOODFREXP-NEXT: [[SQRT_MD_1ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META2]] ; IEEE-GOODFREXP-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 0 ; IEEE-GOODFREXP-NEXT: [[TMP2:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 1 ; IEEE-GOODFREXP-NEXT: [[TMP3:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -2136,7 +2136,7 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; IEEE-GOODFREXP-NEXT: [[TMP17:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i64 0 ; IEEE-GOODFREXP-NEXT: [[MD_1ULP:%.*]] = insertelement <2 x float> [[TMP17]], float [[TMP16]], i64 1 ; IEEE-GOODFREXP-NEXT: store volatile <2 x float> [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[SQRT_MD_1ULP_UNDEF:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath !2 +; IEEE-GOODFREXP-NEXT: [[SQRT_MD_1ULP_UNDEF:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META2]] ; IEEE-GOODFREXP-NEXT: [[TMP18:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP_UNDEF]], i64 0 ; IEEE-GOODFREXP-NEXT: [[TMP19:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP_UNDEF]], i64 1 ; IEEE-GOODFREXP-NEXT: [[TMP20:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -2160,7 +2160,7 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; IEEE-GOODFREXP-NEXT: [[TMP38:%.*]] = insertelement <2 x float> poison, float [[TMP27]], i64 0 ; IEEE-GOODFREXP-NEXT: [[MD_1ULP_UNDEF:%.*]] = insertelement <2 x float> [[TMP38]], float [[TMP37]], i64 1 ; IEEE-GOODFREXP-NEXT: store volatile <2 x float> [[MD_1ULP_UNDEF]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath !3 +; IEEE-GOODFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META3:![0-9]+]] ; IEEE-GOODFREXP-NEXT: [[TMP39:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 0 ; IEEE-GOODFREXP-NEXT: [[TMP40:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 1 ; IEEE-GOODFREXP-NEXT: [[TMP41:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -2187,7 +2187,7 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; IEEE-BADFREXP-NEXT: [[SQRT_X_NO_MD:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]) ; IEEE-BADFREXP-NEXT: [[NO_MD:%.*]] = fdiv contract <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[SQRT_X_NO_MD]] ; IEEE-BADFREXP-NEXT: store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[SQRT_MD_1ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath !2 +; IEEE-BADFREXP-NEXT: [[SQRT_MD_1ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META2]] ; IEEE-BADFREXP-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 0 ; IEEE-BADFREXP-NEXT: [[TMP2:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 1 ; IEEE-BADFREXP-NEXT: [[TMP3:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -2207,7 +2207,7 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; IEEE-BADFREXP-NEXT: [[TMP17:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i64 0 ; IEEE-BADFREXP-NEXT: [[MD_1ULP:%.*]] = insertelement <2 x float> [[TMP17]], float [[TMP16]], i64 1 ; IEEE-BADFREXP-NEXT: store volatile <2 x float> [[MD_1ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[SQRT_MD_1ULP_UNDEF:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath !2 +; IEEE-BADFREXP-NEXT: [[SQRT_MD_1ULP_UNDEF:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META2]] ; IEEE-BADFREXP-NEXT: [[TMP18:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP_UNDEF]], i64 0 ; IEEE-BADFREXP-NEXT: [[TMP19:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP_UNDEF]], i64 1 ; IEEE-BADFREXP-NEXT: [[TMP20:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -2231,7 +2231,7 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; IEEE-BADFREXP-NEXT: [[TMP38:%.*]] = insertelement <2 x float> poison, float [[TMP27]], i64 0 ; IEEE-BADFREXP-NEXT: [[MD_1ULP_UNDEF:%.*]] = insertelement <2 x float> [[TMP38]], float [[TMP37]], i64 1 ; IEEE-BADFREXP-NEXT: store volatile <2 x float> [[MD_1ULP_UNDEF]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath !3 +; IEEE-BADFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META3:![0-9]+]] ; IEEE-BADFREXP-NEXT: [[TMP39:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 0 ; IEEE-BADFREXP-NEXT: [[TMP40:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 1 ; IEEE-BADFREXP-NEXT: [[TMP41:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -2258,7 +2258,7 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; DAZ-NEXT: [[SQRT_X_NO_MD:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]) ; DAZ-NEXT: [[NO_MD:%.*]] = fdiv contract <2 x float> <float 1.000000e+00, float 1.000000e+00>, [[SQRT_X_NO_MD]] ; DAZ-NEXT: store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[SQRT_MD_1ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath !2 +; DAZ-NEXT: [[SQRT_MD_1ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META2:![0-9]+]] ; DAZ-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 0 ; DAZ-NEXT: [[TMP2:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 1 ; DAZ-NEXT: [[TMP3:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -2290,7 +2290,7 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; DAZ-NEXT: [[TMP26:%.*]] = insertelement <2 x float> poison, float [[TMP15]], i64 0 ; DAZ-NEXT: [[MD_1ULP_UNDEF:%.*]] = insertelement <2 x float> [[TMP26]], float [[TMP25]], i64 1 ; DAZ-NEXT: store volatile <2 x float> [[MD_1ULP_UNDEF]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[SQRT_X_3ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath !3 +; DAZ-NEXT: [[SQRT_X_3ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META3:![0-9]+]] ; DAZ-NEXT: [[TMP27:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 0 ; DAZ-NEXT: [[TMP28:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 1 ; DAZ-NEXT: [[TMP29:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -3086,7 +3086,7 @@ define amdgpu_kernel void @multiple_arcp_fdiv_sqrt_denom_25ulp_x3(ptr addrspace( define <4 x float> @rsq_f32_vector_mixed_constant_numerator(<4 x float> %arg) { ; IEEE-GOODFREXP-LABEL: define <4 x float> @rsq_f32_vector_mixed_constant_numerator( ; IEEE-GOODFREXP-SAME: <4 x float> [[ARG:%.*]]) #[[ATTR1]] { -; IEEE-GOODFREXP-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath !2 +; IEEE-GOODFREXP-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath [[META2]] ; IEEE-GOODFREXP-NEXT: [[TMP1:%.*]] = extractelement <4 x float> [[DENOM]], i64 0 ; IEEE-GOODFREXP-NEXT: [[TMP2:%.*]] = extractelement <4 x float> [[DENOM]], i64 1 ; IEEE-GOODFREXP-NEXT: [[TMP3:%.*]] = extractelement <4 x float> [[DENOM]], i64 2 @@ -3135,7 +3135,7 @@ define <4 x float> @rsq_f32_vector_mixed_constant_numerator(<4 x float> %arg) { ; ; IEEE-BADFREXP-LABEL: define <4 x float> @rsq_f32_vector_mixed_constant_numerator( ; IEEE-BADFREXP-SAME: <4 x float> [[ARG:%.*]]) #[[ATTR1]] { -; IEEE-BADFREXP-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath !2 +; IEEE-BADFREXP-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath [[META2]] ; IEEE-BADFREXP-NEXT: [[TMP1:%.*]] = extractelement <4 x float> [[DENOM]], i64 0 ; IEEE-BADFREXP-NEXT: [[TMP2:%.*]] = extractelement <4 x float> [[DENOM]], i64 1 ; IEEE-BADFREXP-NEXT: [[TMP3:%.*]] = extractelement <4 x float> [[DENOM]], i64 2 @@ -3381,7 +3381,7 @@ define <4 x float> @rsq_f32_vector_mixed_constant_numerator_afn_sqrt(<4 x float> define <4 x float> @rsq_f32_vector_mixed_constant_numerator_afn_div(<4 x float> %arg) { ; IEEE-LABEL: define <4 x float> @rsq_f32_vector_mixed_constant_numerator_afn_div( ; IEEE-SAME: <4 x float> [[ARG:%.*]]) #[[ATTR1]] { -; IEEE-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath !2 +; IEEE-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath [[META2:![0-9]+]] ; IEEE-NEXT: [[PARTIAL_RSQ:%.*]] = fdiv contract afn <4 x float> <float 1.000000e+00, float -1.000000e+00, float 4.000000e+00, float undef>, [[DENOM]] ; IEEE-NEXT: ret <4 x float> [[PARTIAL_RSQ]] ; @@ -3410,7 +3410,7 @@ define <4 x float> @rsq_f32_vector_mixed_constant_numerator_afn_div(<4 x float> define <4 x float> @rsq_f32_vector_mixed_constant_numerator_correct_fdiv(<4 x float> %arg) { ; IEEE-LABEL: define <4 x float> @rsq_f32_vector_mixed_constant_numerator_correct_fdiv( ; IEEE-SAME: <4 x float> [[ARG:%.*]]) #[[ATTR1]] { -; IEEE-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath !2 +; IEEE-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath [[META2]] ; IEEE-NEXT: [[PARTIAL_RSQ:%.*]] = fdiv contract <4 x float> <float 1.000000e+00, float -1.000000e+00, float 4.000000e+00, float undef>, [[DENOM]] ; IEEE-NEXT: ret <4 x float> [[PARTIAL_RSQ]] ; @@ -3573,7 +3573,7 @@ define <4 x float> @rsq_f32_vector_mixed_constant_numerator_correct_sqrt(<4 x fl define <4 x float> @rsq_f32_vector_mixed_constant_numerator_arcp(<4 x float> %arg) { ; IEEE-GOODFREXP-LABEL: define <4 x float> @rsq_f32_vector_mixed_constant_numerator_arcp( ; IEEE-GOODFREXP-SAME: <4 x float> [[ARG:%.*]]) #[[ATTR1]] { -; IEEE-GOODFREXP-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath !2 +; IEEE-GOODFREXP-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath [[META2]] ; IEEE-GOODFREXP-NEXT: [[TMP1:%.*]] = extractelement <4 x float> [[DENOM]], i64 0 ; IEEE-GOODFREXP-NEXT: [[TMP2:%.*]] = extractelement <4 x float> [[DENOM]], i64 1 ; IEEE-GOODFREXP-NEXT: [[TMP3:%.*]] = extractelement <4 x float> [[DENOM]], i64 2 @@ -3616,7 +3616,7 @@ define <4 x float> @rsq_f32_vector_mixed_constant_numerator_arcp(<4 x float> %ar ; ; IEEE-BADFREXP-LABEL: define <4 x float> @rsq_f32_vector_mixed_constant_numerator_arcp( ; IEEE-BADFREXP-SAME: <4 x float> [[ARG:%.*]]) #[[ATTR1]] { -; IEEE-BADFREXP-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath !2 +; IEEE-BADFREXP-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath [[META2]] ; IEEE-BADFREXP-NEXT: [[TMP1:%.*]] = extractelement <4 x float> [[DENOM]], i64 0 ; IEEE-BADFREXP-NEXT: [[TMP2:%.*]] = extractelement <4 x float> [[DENOM]], i64 1 ; IEEE-BADFREXP-NEXT: [[TMP3:%.*]] = extractelement <4 x float> [[DENOM]], i64 2 @@ -3696,7 +3696,7 @@ define <4 x float> @rsq_f32_vector_mixed_constant_numerator_arcp(<4 x float> %ar define <4 x float> @rsq_f32_vector_mixed_constant_numerator_arcp_correct(<4 x float> %arg) { ; IEEE-LABEL: define <4 x float> @rsq_f32_vector_mixed_constant_numerator_arcp_correct( ; IEEE-SAME: <4 x float> [[ARG:%.*]]) #[[ATTR1]] { -; IEEE-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath !2 +; IEEE-NEXT: [[DENOM:%.*]] = call contract <4 x float> @llvm.sqrt.v4f32(<4 x float> [[ARG]]), !fpmath [[META2]] ; IEEE-NEXT: [[PARTIAL_RSQ:%.*]] = fdiv arcp contract <4 x float> <float 1.000000e+00, float -1.000000e+00, float 4.000000e+00, float undef>, [[DENOM]] ; IEEE-NEXT: ret <4 x float> [[PARTIAL_RSQ]] ; @@ -3857,21 +3857,21 @@ define <4 x float> @rsq_f32_vector_const_denom(ptr addrspace(1) %out, <2 x float ; IEEE-GOODFREXP-NEXT: [[TMP16:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP13]]) ; IEEE-GOODFREXP-NEXT: [[TMP17:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP16]], i32 [[TMP15]]) ; IEEE-GOODFREXP-NEXT: [[TMP18:%.*]] = fneg contract float [[TMP9]] -; IEEE-GOODFREXP-NEXT: [[TMP19:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP18]]) -; IEEE-GOODFREXP-NEXT: [[TMP20:%.*]] = extractvalue { float, i32 } [[TMP19]], 0 -; IEEE-GOODFREXP-NEXT: [[TMP21:%.*]] = extractvalue { float, i32 } [[TMP19]], 1 -; IEEE-GOODFREXP-NEXT: [[TMP22:%.*]] = sub i32 0, [[TMP21]] -; IEEE-GOODFREXP-NEXT: [[TMP23:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP20]]) -; IEEE-GOODFREXP-NEXT: [[TMP24:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP23]], i32 [[TMP22]]) -; IEEE-GOODFREXP-NEXT: [[TMP25:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP10]]) +; IEEE-GOODFREXP-NEXT: [[TMP25:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP18]]) ; IEEE-GOODFREXP-NEXT: [[TMP26:%.*]] = extractvalue { float, i32 } [[TMP25]], 0 ; IEEE-GOODFREXP-NEXT: [[TMP27:%.*]] = extractvalue { float, i32 } [[TMP25]], 1 +; IEEE-GOODFREXP-NEXT: [[TMP22:%.*]] = sub i32 0, [[TMP27]] ; IEEE-GOODFREXP-NEXT: [[TMP28:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP26]]) +; IEEE-GOODFREXP-NEXT: [[TMP24:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP28]], i32 [[TMP22]]) +; IEEE-GOODFREXP-NEXT: [[TMP48:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP10]]) +; IEEE-GOODFREXP-NEXT: [[TMP49:%.*]] = extractvalue { float, i32 } [[TMP48]], 0 +; IEEE-GOODFREXP-NEXT: [[TMP50:%.*]] = extractvalue { float, i32 } [[TMP48]], 1 +; IEEE-GOODFREXP-NEXT: [[TMP51:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP49]]) ; IEEE-GOODFREXP-NEXT: [[TMP29:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float undef) ; IEEE-GOODFREXP-NEXT: [[TMP30:%.*]] = extractvalue { float, i32 } [[TMP29]], 0 ; IEEE-GOODFREXP-NEXT: [[TMP31:%.*]] = extractvalue { float, i32 } [[TMP29]], 1 -; IEEE-GOODFREXP-NEXT: [[TMP32:%.*]] = fmul contract float [[TMP30]], [[TMP28]] -; IEEE-GOODFREXP-NEXT: [[TMP33:%.*]] = sub i32 [[TMP31]], [[TMP27]] +; IEEE-GOODFREXP-NEXT: [[TMP32:%.*]] = fmul contract float [[TMP30]], [[TMP51]] +; IEEE-GOODFREXP-NEXT: [[TMP33:%.*]] = sub i32 [[TMP31]], [[TMP50]] ; IEEE-GOODFREXP-NEXT: [[TMP34:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP32]], i32 [[TMP33]]) ; IEEE-GOODFREXP-NEXT: [[TMP35:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP11]]) ; IEEE-GOODFREXP-NEXT: [[TMP36:%.*]] = extractvalue { float, i32 } [[TMP35]], 0 @@ -3910,20 +3910,20 @@ define <4 x float> @rsq_f32_vector_const_denom(ptr addrspace(1) %out, <2 x float ; IEEE-BADFREXP-NEXT: [[TMP16:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP13]]) ; IEEE-BADFREXP-NEXT: [[TMP17:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP16]], i32 [[TMP15]]) ; IEEE-BADFREXP-NEXT: [[TMP18:%.*]] = fneg contract float [[TMP9]] -; IEEE-BADFREXP-NEXT: [[TMP19:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP18]]) -; IEEE-BADFREXP-NEXT: [[TMP20:%.*]] = extractvalue { float, i32 } [[TMP19]], 0 +; IEEE-BADFREXP-NEXT: [[TMP25:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP18]]) +; IEEE-BADFREXP-NEXT: [[TMP26:%.*]] = extractvalue { float, i32 } [[TMP25]], 0 ; IEEE-BADFREXP-NEXT: [[TMP21:%.*]] = call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[TMP18]]) ; IEEE-BADFREXP-NEXT: [[TMP22:%.*]] = sub i32 0, [[TMP21]] -; IEEE-BADFREXP-NEXT: [[TMP23:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP20]]) -; IEEE-BADFREXP-NEXT: [[TMP24:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP23]], i32 [[TMP22]]) -; IEEE-BADFREXP-NEXT: [[TMP25:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP10]]) -; IEEE-BADFREXP-NEXT: [[TMP26:%.*]] = extractvalue { float, i32 } [[TMP25]], 0 -; IEEE-BADFREXP-NEXT: [[TMP27:%.*]] = call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[TMP10]]) ; IEEE-BADFREXP-NEXT: [[TMP28:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP26]]) +; IEEE-BADFREXP-NEXT: [[TMP24:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP28]], i32 [[TMP22]]) +; IEEE-BADFREXP-NEXT: [[TMP48:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP10]]) +; IEEE-BADFREXP-NEXT: [[TMP49:%.*]] = extractvalue { float, i32 } [[TMP48]], 0 +; IEEE-BADFREXP-NEXT: [[TMP27:%.*]] = call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[TMP10]]) +; IEEE-BADFREXP-NEXT: [[TMP50:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP49]]) ; IEEE-BADFREXP-NEXT: [[TMP29:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float undef) ; IEEE-BADFREXP-NEXT: [[TMP30:%.*]] = extractvalue { float, i32 } [[TMP29]], 0 ; IEEE-BADFREXP-NEXT: [[TMP31:%.*]] = call i32 @llvm.amdgcn.frexp.exp.i32.f32(float undef) -; IEEE-BADFREXP-NEXT: [[TMP32:%.*]] = fmul contract float [[TMP30]], [[TMP28]] +; IEEE-BADFREXP-NEXT: [[TMP32:%.*]] = fmul contract float [[TMP30]], [[TMP50]] ; IEEE-BADFREXP-NEXT: [[TMP33:%.*]] = sub i32 [[TMP31]], [[TMP27]] ; IEEE-BADFREXP-NEXT: [[TMP34:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP32]], i32 [[TMP33]]) ; IEEE-BADFREXP-NEXT: [[TMP35:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP11]]) @@ -4110,7 +4110,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_lhs(ptr addrspace(1) %out, floa ; IEEE-GOODFREXP-SAME: ptr addrspace(1) [[OUT:%.*]], float nofpclass(sub) [[A:%.*]], float [[B:%.*]]) #[[ATTR1]] { ; IEEE-GOODFREXP-NEXT: [[NO_MD:%.*]] = fdiv float [[A]], [[B]] ; IEEE-GOODFREXP-NEXT: store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath !1 +; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1]] ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[B]]) ; IEEE-GOODFREXP-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 @@ -4145,9 +4145,9 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_lhs(ptr addrspace(1) %out, floa ; IEEE-GOODFREXP-NEXT: [[TMP27:%.*]] = sub i32 [[TMP25]], [[TMP21]] ; IEEE-GOODFREXP-NEXT: [[MD_3ULP:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP26]], i32 [[TMP27]]) ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath !0 +; IEEE-GOODFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath [[META0]] ; IEEE-GOODFREXP-NEXT: store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath !0 +; IEEE-GOODFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath [[META0]] ; IEEE-GOODFREXP-NEXT: store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[NO_MD_ARCP:%.*]] = fdiv arcp float [[A]], [[B]] ; IEEE-GOODFREXP-NEXT: store volatile float [[NO_MD_ARCP]], ptr addrspace(1) [[OUT]], align 4 @@ -4173,7 +4173,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_lhs(ptr addrspace(1) %out, floa ; IEEE-BADFREXP-SAME: ptr addrspace(1) [[OUT:%.*]], float nofpclass(sub) [[A:%.*]], float [[B:%.*]]) #[[ATTR1]] { ; IEEE-BADFREXP-NEXT: [[NO_MD:%.*]] = fdiv float [[A]], [[B]] ; IEEE-BADFREXP-NEXT: store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath !1 +; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1]] ; IEEE-BADFREXP-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[B]]) ; IEEE-BADFREXP-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 @@ -4208,9 +4208,9 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_lhs(ptr addrspace(1) %out, floa ; IEEE-BADFREXP-NEXT: [[TMP27:%.*]] = sub i32 [[TMP25]], [[TMP21]] ; IEEE-BADFREXP-NEXT: [[MD_3ULP:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP26]], i32 [[TMP27]]) ; IEEE-BADFREXP-NEXT: store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath !0 +; IEEE-BADFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath [[META0]] ; IEEE-BADFREXP-NEXT: store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath !0 +; IEEE-BADFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath [[META0]] ; IEEE-BADFREXP-NEXT: store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[NO_MD_ARCP:%.*]] = fdiv arcp float [[A]], [[B]] ; IEEE-BADFREXP-NEXT: store volatile float [[NO_MD_ARCP]], ptr addrspace(1) [[OUT]], align 4 @@ -4236,7 +4236,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_lhs(ptr addrspace(1) %out, floa ; DAZ-SAME: ptr addrspace(1) [[OUT:%.*]], float nofpclass(sub) [[A:%.*]], float [[B:%.*]]) #[[ATTR1]] { ; DAZ-NEXT: [[NO_MD:%.*]] = fdiv float [[A]], [[B]] ; DAZ-NEXT: store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath !1 +; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1]] ; DAZ-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[B]]) ; DAZ-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 @@ -4253,9 +4253,9 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_lhs(ptr addrspace(1) %out, floa ; DAZ-NEXT: store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[MD_3ULP:%.*]] = call float @llvm.amdgcn.fdiv.fast(float [[A]], float [[B]]) ; DAZ-NEXT: store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath !0 +; DAZ-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath [[META0]] ; DAZ-NEXT: store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath !0 +; DAZ-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath [[META0]] ; DAZ-NEXT: store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[NO_MD_ARCP:%.*]] = fdiv arcp float [[A]], [[B]] ; DAZ-NEXT: store volatile float [[NO_MD_ARCP]], ptr addrspace(1) [[OUT]], align 4 @@ -4295,7 +4295,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_rhs(ptr addrspace(1) %out, floa ; IEEE-GOODFREXP-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float nofpclass(sub) [[B:%.*]]) #[[ATTR1]] { ; IEEE-GOODFREXP-NEXT: [[NO_MD:%.*]] = fdiv float [[A]], [[B]] ; IEEE-GOODFREXP-NEXT: store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath !1 +; IEEE-GOODFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1]] ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[B]]) ; IEEE-GOODFREXP-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 @@ -4330,9 +4330,9 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_rhs(ptr addrspace(1) %out, floa ; IEEE-GOODFREXP-NEXT: [[TMP27:%.*]] = sub i32 [[TMP25]], [[TMP21]] ; IEEE-GOODFREXP-NEXT: [[MD_3ULP:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP26]], i32 [[TMP27]]) ; IEEE-GOODFREXP-NEXT: store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath !0 +; IEEE-GOODFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath [[META0]] ; IEEE-GOODFREXP-NEXT: store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath !0 +; IEEE-GOODFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath [[META0]] ; IEEE-GOODFREXP-NEXT: store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-GOODFREXP-NEXT: [[NO_MD_ARCP:%.*]] = fdiv arcp float [[A]], [[B]] ; IEEE-GOODFREXP-NEXT: store volatile float [[NO_MD_ARCP]], ptr addrspace(1) [[OUT]], align 4 @@ -4358,7 +4358,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_rhs(ptr addrspace(1) %out, floa ; IEEE-BADFREXP-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float nofpclass(sub) [[B:%.*]]) #[[ATTR1]] { ; IEEE-BADFREXP-NEXT: [[NO_MD:%.*]] = fdiv float [[A]], [[B]] ; IEEE-BADFREXP-NEXT: store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath !1 +; IEEE-BADFREXP-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1]] ; IEEE-BADFREXP-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[B]]) ; IEEE-BADFREXP-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 @@ -4393,9 +4393,9 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_rhs(ptr addrspace(1) %out, floa ; IEEE-BADFREXP-NEXT: [[TMP27:%.*]] = sub i32 [[TMP25]], [[TMP21]] ; IEEE-BADFREXP-NEXT: [[MD_3ULP:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP26]], i32 [[TMP27]]) ; IEEE-BADFREXP-NEXT: store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath !0 +; IEEE-BADFREXP-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath [[META0]] ; IEEE-BADFREXP-NEXT: store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath !0 +; IEEE-BADFREXP-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath [[META0]] ; IEEE-BADFREXP-NEXT: store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; IEEE-BADFREXP-NEXT: [[NO_MD_ARCP:%.*]] = fdiv arcp float [[A]], [[B]] ; IEEE-BADFREXP-NEXT: store volatile float [[NO_MD_ARCP]], ptr addrspace(1) [[OUT]], align 4 @@ -4421,7 +4421,7 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_rhs(ptr addrspace(1) %out, floa ; DAZ-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float nofpclass(sub) [[B:%.*]]) #[[ATTR1]] { ; DAZ-NEXT: [[NO_MD:%.*]] = fdiv float [[A]], [[B]] ; DAZ-NEXT: store volatile float [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath !1 +; DAZ-NEXT: [[MD_HALF_ULP:%.*]] = fdiv float [[A]], [[B]], !fpmath [[META1]] ; DAZ-NEXT: store volatile float [[MD_HALF_ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[TMP1:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[B]]) ; DAZ-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP1]], 0 @@ -4438,9 +4438,9 @@ define amdgpu_kernel void @fdiv_fpmath_f32_nosub_rhs(ptr addrspace(1) %out, floa ; DAZ-NEXT: store volatile float [[MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[MD_3ULP:%.*]] = call float @llvm.amdgcn.fdiv.fast(float [[A]], float [[B]]) ; DAZ-NEXT: store volatile float [[MD_3ULP]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath !0 +; DAZ-NEXT: [[FAST_MD_25ULP:%.*]] = fdiv fast float [[A]], [[B]], !fpmath [[META0]] ; DAZ-NEXT: store volatile float [[FAST_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath !0 +; DAZ-NEXT: [[AFN_MD_25ULP:%.*]] = fdiv afn float [[A]], [[B]], !fpmath [[META0]] ; DAZ-NEXT: store volatile float [[AFN_MD_25ULP]], ptr addrspace(1) [[OUT]], align 4 ; DAZ-NEXT: [[NO_MD_ARCP:%.*]] = fdiv arcp float [[A]], [[B]] ; DAZ-NEXT: store volatile float [[NO_MD_ARCP]], ptr addrspace(1) [[OUT]], align 4 |