; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -stop-after=si-fix-sgpr-copies -verify-machineinstrs < %s | FileCheck %s
define amdgpu_kernel void @exp_f32(ptr addrspace(1) %ptr) {
; CHECK-LABEL: name: exp_f32
; CHECK: bb.0 (%ir-block.0):
; CHECK-NEXT: liveins: $sgpr2_sgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr2_sgpr3
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec_xnull = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64) from %ir.ptr.kernarg.offset, align 4, addrspace 4)
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s32) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_EXP_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_EXP_F32_e64 0, [[GLOBAL_LOAD_DWORD_SADDR]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[V_EXP_F32_e64_]]
; CHECK-NEXT: GLOBAL_STORE_DWORD_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s32) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
%val = load volatile float, ptr addrspace(1) %ptr
%res = call float @llvm.amdgcn.exp2.f32(float %val)
store float %res, ptr addrspace(1) %ptr
ret void
}
define amdgpu_kernel void @exp_f16(ptr addrspace(1) %ptr) {
; CHECK-LABEL: name: exp_f16
; CHECK: bb.0 (%ir-block.0):
; CHECK-NEXT: liveins: $sgpr2_sgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr2_sgpr3
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec_xnull = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64) from %ir.ptr.kernarg.offset, align 4, addrspace 4)
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_USHORT_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_USHORT_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s16) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_EXP_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_EXP_F16_fake16_e64 0, [[GLOBAL_LOAD_USHORT_SADDR]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[DEF1:%[0-9]+]]:sreg_32_xexec = IMPLICIT_DEF
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[V_EXP_F16_fake16_e64_]]
; CHECK-NEXT: GLOBAL_STORE_SHORT_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s16) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
%val = load volatile half, ptr addrspace(1) %ptr
%res = call half @llvm.amdgcn.exp2.f16(half %val)
store half %res, ptr addrspace(1) %ptr
ret void
}
define amdgpu_kernel void @log_f32(ptr addrspace(1) %ptr) {
; CHECK-LABEL: name: log_f32
; CHECK: bb.0 (%ir-block.0):
; CHECK-NEXT: liveins: $sgpr2_sgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr2_sgpr3
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec_xnull = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64) from %ir.ptr.kernarg.offset, align 4, addrspace 4)
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s32) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_LOG_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_LOG_F32_e64 0, [[GLOBAL_LOAD_DWORD_SADDR]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[V_LOG_F32_e64_]]
; CHECK-NEXT: GLOBAL_STORE_DWORD_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s32) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
%val = load volatile float, ptr addrspace(1) %ptr
%res = call float @llvm.amdgcn.log.f32(float %val)
store float %res, ptr addrspace(1) %ptr
ret void
}
define amdgpu_kernel void @log_f16(ptr addrspace(1) %ptr) {
; CHECK-LABEL: name: log_f16
; CHECK: bb.0 (%ir-block.0):
; CHECK-NEXT: liveins: $sgpr2_sgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr2_sgpr3
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec_xnull = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64) from %ir.ptr.kernarg.offset, align 4, addrspace 4)
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_USHORT_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_USHORT_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s16) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_LOG_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_LOG_F16_fake16_e64 0, [[GLOBAL_LOAD_USHORT_SADDR]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[DEF1:%[0-9]+]]:sreg_32_xexec = IMPLICIT_DEF
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[V_LOG_F16_fake16_e64_]]
; CHECK-NEXT: GLOBAL_STORE_SHORT_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s16) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
%val = load volatile half, ptr addrspace(1) %ptr
%res = call half @llvm.amdgcn.log.f16(half %val)
store half %res, ptr addrspace(1) %ptr
ret void
}
define amdgpu_kernel void @rcp_f32(ptr addrspace(1) %ptr) {
; CHECK-LABEL: name: rcp_f32
; CHECK: bb.0 (%ir-block.0):
; CHECK-NEXT: liveins: $sgpr2_sgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr2_sgpr3
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec_xnull = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64) from %ir.ptr.kernarg.offset, align 4, addrspace 4)
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s32) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_RCP_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_RCP_F32_e64 0, [[GLOBAL_LOAD_DWORD_SADDR]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[V_RCP_F32_e64_]]
; CHECK-NEXT: GLOBAL_STORE_DWORD_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s32) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
%val = load volatile float, ptr addrspace(1) %ptr
%res = call float @llvm.amdgcn.rcp.f32(float %val)
store float %res, ptr addrspace(1) %ptr
ret void
}
define amdgpu_kernel void @rcp_f16(ptr addrspace(1) %ptr) {
; CHECK-LABEL: name: rcp_f16
; CHECK: bb.0 (%ir-block.0):
; CHECK-NEXT: liveins: $sgpr2_sgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr2_sgpr3
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec_xnull = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64) from %ir.ptr.kernarg.offset, align 4, addrspace 4)
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_USHORT_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_USHORT_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s16) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_RCP_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_RCP_F16_fake16_e64 0, [[GLOBAL_LOAD_USHORT_SADDR]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[DEF1:%[0-9]+]]:sreg_32_xexec = IMPLICIT_DEF
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[V_RCP_F16_fake16_e64_]]
; CHECK-NEXT: GLOBAL_STORE_SHORT_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s16) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
%val = load volatile half, ptr addrspace(1) %ptr
%res = call half @llvm.amdgcn.rcp.f16(half %val)
store half %res, ptr addrspace(1) %ptr
ret void
}
define amdgpu_kernel void @rsq_f32(ptr addrspace(1) %ptr) {
; CHECK-LABEL: name: rsq_f32
; CHECK: bb.0 (%ir-block.0):
; CHECK-NEXT: liveins: $sgpr2_sgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr2_sgpr3
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec_xnull = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64) from %ir.ptr.kernarg.offset, align 4, addrspace 4)
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s32) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_RSQ_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_RSQ_F32_e64 0, [[GLOBAL_LOAD_DWORD_SADDR]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[V_RSQ_F32_e64_]]
; CHECK-NEXT: GLOBAL_STORE_DWORD_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s32) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
%val = load volatile float, ptr addrspace(1) %ptr
%res = call float @llvm.amdgcn.rsq.f32(float %val)
store float %res, ptr addrspace(1) %ptr
ret void
}
define amdgpu_kernel void @rsq_f16(ptr addrspace(1) %ptr) {
; CHECK-LABEL: name: rsq_f16
; CHECK: bb.0 (%ir-block.0):
; CHECK-NEXT: liveins: $sgpr2_sgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr2_sgpr3
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec_xnull = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64) from %ir.ptr.kernarg.offset, align 4, addrspace 4)
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_USHORT_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_USHORT_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s16) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_RSQ_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_RSQ_F16_fake16_e64 0, [[GLOBAL_LOAD_USHORT_SADDR]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[DEF1:%[0-9]+]]:sreg_32_xexec = IMPLICIT_DEF
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[V_RSQ_F16_fake16_e64_]]
; CHECK-NEXT: GLOBAL_STORE_SHORT_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s16) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
%val = load volatile half, ptr addrspace(1) %ptr
%res = call half @llvm.amdgcn.rsq.f16(half %val)
store half %res, ptr addrspace(1) %ptr
ret void
}
define amdgpu_kernel void @sqrt_f32(ptr addrspace(1) %ptr) {
; CHECK-LABEL: name: sqrt_f32
; CHECK: bb.0 (%ir-block.0):
; CHECK-NEXT: liveins: $sgpr2_sgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr2_sgpr3
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec_xnull = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64) from %ir.ptr.kernarg.offset, align 4, addrspace 4)
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s32) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_SQRT_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_SQRT_F32_e64 0, [[GLOBAL_LOAD_DWORD_SADDR]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[V_SQRT_F32_e64_]]
; CHECK-NEXT: GLOBAL_STORE_DWORD_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s32) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
%val = load volatile float, ptr addrspace(1) %ptr
%res = call float @llvm.amdgcn.sqrt.f32(float %val)
store float %res, ptr addrspace(1) %ptr
ret void
}
define amdgpu_kernel void @sqrt_f16(ptr addrspace(1) %ptr) {
; CHECK-LABEL: name: sqrt_f16
; CHECK: bb.0 (%ir-block.0):
; CHECK-NEXT: liveins: $sgpr2_sgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr2_sgpr3
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec_xnull = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64) from %ir.ptr.kernarg.offset, align 4, addrspace 4)
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_USHORT_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_USHORT_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s16) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
; CHECK-NEXT: [[V_SQRT_F16_fake16_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_SQRT_F16_fake16_e64 0, [[GLOBAL_LOAD_USHORT_SADDR]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[DEF1:%[0-9]+]]:sreg_32_xexec = IMPLICIT_DEF
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[V_SQRT_F16_fake16_e64_]]
; CHECK-NEXT: GLOBAL_STORE_SHORT_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s16) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
%val = load volatile half, ptr addrspace(1) %ptr
%res = call half @llvm.amdgcn.sqrt.f16(half %val)
store half %res, ptr addrspace(1) %ptr
ret void
}
declare float @llvm.amdgcn.exp2.f32(float)
declare half @llvm.amdgcn.exp2.f16(half)
declare float @llvm.amdgcn.log.f32(float)
declare half @llvm.amdgcn.log.f16(half)
declare float @llvm.amdgcn.rcp.f32(float)
declare half @llvm.amdgcn.rcp.f16(half)
declare float @llvm.amdgcn.rsq.f32(float)
declare half @llvm.amdgcn.rsq.f16(half)
declare float @llvm.amdgcn.sqrt.f32(float)
declare half @llvm.amdgcn.sqrt.f16(half)