1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 18:54:02 +01:00
llvm-mirror/test/CodeGen/AMDGPU/fptrunc.f16.ll
Matt Arsenault 248de533d0 AMDGPU: Try to eliminate clearing of high bits of 16-bit instructions
These used to consistently be zeroed pre-gfx9, but gfx9 made the
situation complicated since now some still do and some don't. This
also manages to pick up a few cases that the pattern fails to optimize
away.

We handle some cases with instruction patterns, but some get
through. In particular this improves the integer cases.
2021-06-22 13:42:49 -04:00

197 lines
7.2 KiB
LLVM

; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=SI -check-prefix=SIVI %s
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=VI -check-prefix=SIVI %s
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -denormal-fp-math=preserve-sign -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -enable-var-scope -check-prefix=GCN -check-prefix=GFX9 %s
; GCN-LABEL: {{^}}fptrunc_f32_to_f16:
; GCN: buffer_load_dword v[[A_F32:[0-9]+]]
; GCN: v_cvt_f16_f32_e32 v[[R_F16:[0-9]+]], v[[A_F32]]
; GCN: buffer_store_short v[[R_F16]]
; GCN: s_endpgm
define amdgpu_kernel void @fptrunc_f32_to_f16(
half addrspace(1)* %r,
float addrspace(1)* %a) {
entry:
%a.val = load float, float addrspace(1)* %a
%r.val = fptrunc float %a.val to half
store half %r.val, half addrspace(1)* %r
ret void
}
; GCN-LABEL: {{^}}fptrunc_f64_to_f16:
; GCN: buffer_load_dwordx2 v{{\[}}[[A_F64_0:[0-9]+]]:[[A_F64_1:[0-9]+]]{{\]}}
; GCN: v_cvt_f32_f64_e32 v[[A_F32:[0-9]+]], v{{\[}}[[A_F64_0]]:[[A_F64_1]]{{\]}}
; GCN: v_cvt_f16_f32_e32 v[[R_F16:[0-9]+]], v[[A_F32]]
; GCN: buffer_store_short v[[R_F16]]
; GCN: s_endpgm
define amdgpu_kernel void @fptrunc_f64_to_f16(
half addrspace(1)* %r,
double addrspace(1)* %a) {
entry:
%a.val = load double, double addrspace(1)* %a
%r.val = fptrunc double %a.val to half
store half %r.val, half addrspace(1)* %r
ret void
}
; GCN-LABEL: {{^}}fptrunc_v2f32_to_v2f16:
; GCN: buffer_load_dwordx2 v{{\[}}[[A_F32_0:[0-9]+]]:[[A_F32_1:[0-9]+]]{{\]}}
; GCN-DAG: v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[A_F32_0]]
; SI-DAG: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[A_F32_1]]
; SI-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
; SI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_0]], v[[R_F16_HI]]
; VI-DAG: v_cvt_f16_f32_sdwa v[[R_F16_1:[0-9]+]], v[[A_F32_1]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_0]], v[[R_F16_1]]
; GFX9-DAG: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[A_F32_1]]
; GFX9: v_pack_b32_f16 v[[R_V2_F16:[0-9]+]], v[[R_F16_0]], v[[R_F16_1]]
; GCN: buffer_store_dword v[[R_V2_F16]]
; GCN: s_endpgm
define amdgpu_kernel void @fptrunc_v2f32_to_v2f16(
<2 x half> addrspace(1)* %r,
<2 x float> addrspace(1)* %a) {
entry:
%a.val = load <2 x float>, <2 x float> addrspace(1)* %a
%r.val = fptrunc <2 x float> %a.val to <2 x half>
store <2 x half> %r.val, <2 x half> addrspace(1)* %r
ret void
}
; GCN-LABEL: {{^}}fptrunc_v2f64_to_v2f16:
; GCN: buffer_load_dwordx4 v{{\[}}[[A_F64_0:[0-9]+]]:[[A_F64_3:[0-9]+]]{{\]}}
; GCN-DAG: v_cvt_f32_f64_e32 v[[A_F32_0:[0-9]+]], v{{\[}}[[A_F64_0]]:{{[0-9]+}}{{\]}}
; GCN-DAG: v_cvt_f32_f64_e32 v[[A_F32_1:[0-9]+]], v{{\[}}{{[0-9]+}}:[[A_F64_3]]{{\]}}
; GCN-DAG: v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[A_F32_0]]
;
; SI-DAG: v_cvt_f16_f32_e32 v[[CVTHI:[0-9]+]], v[[A_F32_1]]
; SI-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[CVTHI]]
; VI: v_cvt_f16_f32_sdwa v[[R_F16_HI:[0-9]+]], v[[A_F32_1]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; SIVI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_0]], v[[R_F16_HI]]
; GFX9-DAG: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[A_F32_1]]
; GFX9: v_lshl_or_b32 v[[R_V2_F16:[0-9]+]], v[[R_F16_1]], 16, v[[R_F16_0]]
; GCN: buffer_store_dword v[[R_V2_F16]]
define amdgpu_kernel void @fptrunc_v2f64_to_v2f16(
<2 x half> addrspace(1)* %r,
<2 x double> addrspace(1)* %a) {
entry:
%a.val = load <2 x double>, <2 x double> addrspace(1)* %a
%r.val = fptrunc <2 x double> %a.val to <2 x half>
store <2 x half> %r.val, <2 x half> addrspace(1)* %r
ret void
}
; GCN-LABEL: {{^}}fneg_fptrunc_f32_to_f16:
; GCN: buffer_load_dword v[[A_F32:[0-9]+]]
; GCN: v_cvt_f16_f32_e64 v[[R_F16:[0-9]+]], -v[[A_F32]]
; GCN: buffer_store_short v[[R_F16]]
; GCN: s_endpgm
define amdgpu_kernel void @fneg_fptrunc_f32_to_f16(
half addrspace(1)* %r,
float addrspace(1)* %a) {
entry:
%a.val = load float, float addrspace(1)* %a
%a.fneg = fneg float %a.val
%r.val = fptrunc float %a.fneg to half
store half %r.val, half addrspace(1)* %r
ret void
}
; GCN-LABEL: {{^}}fabs_fptrunc_f32_to_f16:
; GCN: buffer_load_dword v[[A_F32:[0-9]+]]
; GCN: v_cvt_f16_f32_e64 v[[R_F16:[0-9]+]], |v[[A_F32]]|
; GCN: buffer_store_short v[[R_F16]]
; GCN: s_endpgm
define amdgpu_kernel void @fabs_fptrunc_f32_to_f16(
half addrspace(1)* %r,
float addrspace(1)* %a) {
entry:
%a.val = load float, float addrspace(1)* %a
%a.fabs = call float @llvm.fabs.f32(float %a.val)
%r.val = fptrunc float %a.fabs to half
store half %r.val, half addrspace(1)* %r
ret void
}
; GCN-LABEL: {{^}}fneg_fabs_fptrunc_f32_to_f16:
; GCN: buffer_load_dword v[[A_F32:[0-9]+]]
; GCN: v_cvt_f16_f32_e64 v[[R_F16:[0-9]+]], -|v[[A_F32]]|
; GCN: buffer_store_short v[[R_F16]]
; GCN: s_endpgm
define amdgpu_kernel void @fneg_fabs_fptrunc_f32_to_f16(
half addrspace(1)* %r,
float addrspace(1)* %a) #0 {
entry:
%a.val = load float, float addrspace(1)* %a
%a.fabs = call float @llvm.fabs.f32(float %a.val)
%a.fneg.fabs = fneg float %a.fabs
%r.val = fptrunc float %a.fneg.fabs to half
store half %r.val, half addrspace(1)* %r
ret void
}
; GCN-LABEL: {{^}}fptrunc_f32_to_f16_zext_i32:
; GCN: buffer_load_dword v[[A_F32:[0-9]+]]
; GCN: v_cvt_f16_f32_e32 v[[R_F16:[0-9]+]], v[[A_F32]]
; SIVI-NOT: v[[R_F16]]
; GFX9-NOT: v_and_b32
; GCN: buffer_store_dword v[[R_F16]]
define amdgpu_kernel void @fptrunc_f32_to_f16_zext_i32(
i32 addrspace(1)* %r,
float addrspace(1)* %a) #0 {
entry:
%a.val = load float, float addrspace(1)* %a
%r.val = fptrunc float %a.val to half
%r.i16 = bitcast half %r.val to i16
%zext = zext i16 %r.i16 to i32
store i32 %zext, i32 addrspace(1)* %r
ret void
}
; GCN-LABEL: {{^}}fptrunc_fabs_f32_to_f16_zext_i32:
; GCN: buffer_load_dword v[[A_F32:[0-9]+]]
; GCN: v_cvt_f16_f32_e64 v[[R_F16:[0-9]+]], |v[[A_F32]]|
; SIVI-NOT: v[[R_F16]]
; GFX9-NOT: v_and_b32
; GCN: buffer_store_dword v[[R_F16]]
define amdgpu_kernel void @fptrunc_fabs_f32_to_f16_zext_i32(
i32 addrspace(1)* %r,
float addrspace(1)* %a) #0 {
entry:
%a.val = load float, float addrspace(1)* %a
%a.fabs = call float @llvm.fabs.f32(float %a.val)
%r.val = fptrunc float %a.fabs to half
%r.i16 = bitcast half %r.val to i16
%zext = zext i16 %r.i16 to i32
store i32 %zext, i32 addrspace(1)* %r
ret void
}
; GCN-LABEL: {{^}}fptrunc_f32_to_f16_sext_i32:
; GCN: buffer_load_dword v[[A_F32:[0-9]+]]
; GCN: v_cvt_f16_f32_e32 v[[R_F16:[0-9]+]], v[[A_F32]]
; GCN: v_bfe_i32 v[[R_F16_SEXT:[0-9]+]], v[[R_F16]], 0, 16
; GCN: buffer_store_dword v[[R_F16_SEXT]]
define amdgpu_kernel void @fptrunc_f32_to_f16_sext_i32(
i32 addrspace(1)* %r,
float addrspace(1)* %a) #0 {
entry:
%a.val = load float, float addrspace(1)* %a
%r.val = fptrunc float %a.val to half
%r.i16 = bitcast half %r.val to i16
%zext = sext i16 %r.i16 to i32
store i32 %zext, i32 addrspace(1)* %r
ret void
}
declare float @llvm.fabs.f32(float) #1
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }