1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s 2 3 ; GCN-LABEL: {{^}}global_truncstore_f64_to_f16: 4 ; GCN: s_endpgm 5 define amdgpu_kernel void @global_truncstore_f64_to_f16(half addrspace(1)* %out, double addrspace(1)* %in) #0 { 6 %val = load double, double addrspace(1)* %in 7 %cvt = fptrunc double %val to half 8 store half %cvt, half addrspace(1)* %out 9 ret void 10 } 11 12 ; GCN-LABEL: {{^}}global_truncstore_v2f64_to_v2f16: 13 ; GCN: s_endpgm 14 define amdgpu_kernel void @global_truncstore_v2f64_to_v2f16(<2 x half> addrspace(1)* %out, <2 x double> addrspace(1)* %in) #0 { 15 %val = load <2 x double>, <2 x double> addrspace(1)* %in 16 %cvt = fptrunc <2 x double> %val to <2 x half> 17 store <2 x half> %cvt, <2 x half> addrspace(1)* %out 18 ret void 19 } 20 21 ; GCN-LABEL: {{^}}global_truncstore_v3f64_to_v3f16: 22 ; GCN: s_endpgm 23 define amdgpu_kernel void @global_truncstore_v3f64_to_v3f16(<3 x half> addrspace(1)* %out, <3 x double> addrspace(1)* %in) #0 { 24 %val = load <3 x double>, <3 x double> addrspace(1)* %in 25 %cvt = fptrunc <3 x double> %val to <3 x half> 26 store <3 x half> %cvt, <3 x half> addrspace(1)* %out 27 ret void 28 } 29 30 ; GCN-LABEL: {{^}}global_truncstore_v4f64_to_v4f16: 31 ; GCN: s_endpgm 32 define amdgpu_kernel void @global_truncstore_v4f64_to_v4f16(<4 x half> addrspace(1)* %out, <4 x double> addrspace(1)* %in) #0 { 33 %val = load <4 x double>, <4 x double> addrspace(1)* %in 34 %cvt = fptrunc <4 x double> %val to <4 x half> 35 store <4 x half> %cvt, <4 x half> addrspace(1)* %out 36 ret void 37 } 38 39 ; GCN-LABEL: {{^}}global_truncstore_v8f64_to_v8f16: 40 ; GCN: s_endpgm 41 define amdgpu_kernel void @global_truncstore_v8f64_to_v8f16(<8 x half> addrspace(1)* %out, <8 x double> addrspace(1)* %in) #0 { 42 %val = load <8 x double>, <8 x double> addrspace(1)* %in 43 %cvt = fptrunc <8 x double> %val to <8 x half> 44 store <8 x half> %cvt, <8 x half> addrspace(1)* %out 45 ret void 46 } 47 48 ; GCN-LABEL: {{^}}global_truncstore_v16f64_to_v16f16: 49 ; GCN: s_endpgm 50 define amdgpu_kernel void @global_truncstore_v16f64_to_v16f16(<16 x half> addrspace(1)* %out, <16 x double> addrspace(1)* %in) #0 { 51 %val = load <16 x double>, <16 x double> addrspace(1)* %in 52 %cvt = fptrunc <16 x double> %val to <16 x half> 53 store <16 x half> %cvt, <16 x half> addrspace(1)* %out 54 ret void 55 } 56
