shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.tanh.ll
Shilei Tian aecd44818a
[AMDGPU] Add support for v_tanh_f16 on gfx1250 (#149439) 
Co-authored-by: Mekhanoshin, Stanislav <Stanislav.Mekhanoshin@amd.com>
2025-07-18 00:21:04 -04:00

262 lines
9.9 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 %s -o - | FileCheck -check-prefix=SDAG-REAL16 %s
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 %s -o - | FileCheck -check-prefix=SDAG-FAKE16 %s
; xUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 %s -o - | FileCheck -check-prefix=GISEL-REAL16 %s
; xUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 %s -o - | FileCheck -check-prefix=GISEL-FAKE16 %s
; FIXME: GlobalISel does not work with bf16
declare float @llvm.amdgcn.tanh.f32(float) #0
declare half @llvm.amdgcn.tanh.f16(half) #0
declare bfloat @llvm.amdgcn.tanh.bf16(bfloat) #0
define amdgpu_kernel void @tanh_f32(ptr addrspace(1) %out, float %src) #1 {
; SDAG-REAL16-LABEL: tanh_f32:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_load_b96 s[0:2], s[4:5], 0x0
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
; SDAG-REAL16-NEXT: v_tanh_f32_e32 v0, s2
; SDAG-REAL16-NEXT: global_store_b32 v1, v0, s[0:1]
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_f32:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_load_b96 s[0:2], s[4:5], 0x0
; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; SDAG-FAKE16-NEXT: v_tanh_f32_e32 v0, s2
; SDAG-FAKE16-NEXT: global_store_b32 v1, v0, s[0:1]
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call float @llvm.amdgcn.tanh.f32(float %src) #0
store float %tanh, ptr addrspace(1) %out, align 4
ret void
}
; TODO: Really these should be constant folded
define amdgpu_kernel void @tanh_f32_constant_4.0(ptr addrspace(1) %out) #1 {
; SDAG-REAL16-LABEL: tanh_f32_constant_4.0:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-REAL16-NEXT: v_tanh_f32_e32 v0, 4.0
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
; SDAG-REAL16-NEXT: global_store_b32 v1, v0, s[0:1]
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_f32_constant_4.0:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-FAKE16-NEXT: v_tanh_f32_e32 v0, 4.0
; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; SDAG-FAKE16-NEXT: global_store_b32 v1, v0, s[0:1]
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call float @llvm.amdgcn.tanh.f32(float 4.0) #0
store float %tanh, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @tanh_f32_constant_100.0(ptr addrspace(1) %out) #1 {
; SDAG-REAL16-LABEL: tanh_f32_constant_100.0:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-REAL16-NEXT: v_tanh_f32_e32 v0, 0x42c80000
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
; SDAG-REAL16-NEXT: global_store_b32 v1, v0, s[0:1]
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_f32_constant_100.0:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-FAKE16-NEXT: v_tanh_f32_e32 v0, 0x42c80000
; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; SDAG-FAKE16-NEXT: global_store_b32 v1, v0, s[0:1]
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call float @llvm.amdgcn.tanh.f32(float 100.0) #0
store float %tanh, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @tanh_undef_f32(ptr addrspace(1) %out) #1 {
; SDAG-REAL16-LABEL: tanh_undef_f32:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_undef_f32:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call float @llvm.amdgcn.tanh.f32(float undef)
store float %tanh, ptr addrspace(1) %out, align 4
ret void
}
define amdgpu_kernel void @tanh_f16(ptr addrspace(1) %out, half %src) #1 {
; SDAG-REAL16-LABEL: tanh_f16:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_load_b96 s[0:2], s[4:5], 0x0
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
; SDAG-REAL16-NEXT: v_tanh_f16_e32 v0.l, s2
; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1]
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_f16:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_load_b96 s[0:2], s[4:5], 0x0
; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; SDAG-FAKE16-NEXT: v_tanh_f16_e32 v0, s2
; SDAG-FAKE16-NEXT: global_store_b16 v1, v0, s[0:1]
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call half @llvm.amdgcn.tanh.f16(half %src) #0
store half %tanh, ptr addrspace(1) %out, align 2
ret void
}
define amdgpu_kernel void @tanh_f16_constant_4.0(ptr addrspace(1) %out) #1 {
; SDAG-REAL16-LABEL: tanh_f16_constant_4.0:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-REAL16-NEXT: v_tanh_f16_e32 v0.l, 4.0
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1]
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_f16_constant_4.0:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-FAKE16-NEXT: v_tanh_f16_e32 v0, 4.0
; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; SDAG-FAKE16-NEXT: global_store_b16 v1, v0, s[0:1]
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call half @llvm.amdgcn.tanh.f16(half 4.0) #0
store half %tanh, ptr addrspace(1) %out, align 2
ret void
}
define amdgpu_kernel void @tanh_f16_constant_100.0(ptr addrspace(1) %out) #1 {
; SDAG-REAL16-LABEL: tanh_f16_constant_100.0:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-REAL16-NEXT: v_tanh_f16_e32 v0.l, 0x5640
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1]
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_f16_constant_100.0:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-FAKE16-NEXT: v_tanh_f16_e32 v0, 0x5640
; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; SDAG-FAKE16-NEXT: global_store_b16 v1, v0, s[0:1]
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call half @llvm.amdgcn.tanh.f16(half 100.0) #0
store half %tanh, ptr addrspace(1) %out, align 2
ret void
}
define amdgpu_kernel void @tanh_undef_f16(ptr addrspace(1) %out) #1 {
; SDAG-REAL16-LABEL: tanh_undef_f16:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_undef_f16:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call half @llvm.amdgcn.tanh.f16(half undef)
store half %tanh, ptr addrspace(1) %out, align 2
ret void
}
define amdgpu_kernel void @tanh_bf16(ptr addrspace(1) %out, bfloat %src) #1 {
; SDAG-REAL16-LABEL: tanh_bf16:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_load_b96 s[0:2], s[4:5], 0x0
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
; SDAG-REAL16-NEXT: v_tanh_bf16_e32 v0.l, s2
; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1]
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_bf16:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_load_b96 s[0:2], s[4:5], 0x0
; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; SDAG-FAKE16-NEXT: v_tanh_bf16_e32 v0, s2
; SDAG-FAKE16-NEXT: global_store_b16 v1, v0, s[0:1]
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call bfloat @llvm.amdgcn.tanh.bf16(bfloat %src) #0
store bfloat %tanh, ptr addrspace(1) %out, align 2
ret void
}
define amdgpu_kernel void @tanh_bf16_constant_4(ptr addrspace(1) %out) #1 {
; SDAG-REAL16-LABEL: tanh_bf16_constant_4:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-REAL16-NEXT: v_tanh_bf16_e32 v0.l, 4.0
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1]
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_bf16_constant_4:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-FAKE16-NEXT: v_tanh_bf16_e32 v0, 4.0
; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; SDAG-FAKE16-NEXT: global_store_b16 v1, v0, s[0:1]
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call bfloat @llvm.amdgcn.tanh.bf16(bfloat 4.0) #0
store bfloat %tanh, ptr addrspace(1) %out, align 2
ret void
}
define amdgpu_kernel void @tanh_bf16_constant_100(ptr addrspace(1) %out) #1 {
; SDAG-REAL16-LABEL: tanh_bf16_constant_100:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-REAL16-NEXT: v_tanh_bf16_e32 v0.l, 0x42c8
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1]
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_bf16_constant_100:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0
; SDAG-FAKE16-NEXT: v_tanh_bf16_e32 v0, 0x42c8
; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
; SDAG-FAKE16-NEXT: global_store_b16 v1, v0, s[0:1]
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call bfloat @llvm.amdgcn.tanh.bf16(bfloat 100.0) #0
store bfloat %tanh, ptr addrspace(1) %out, align 2
ret void
}
define amdgpu_kernel void @tanh_undef_bf16(ptr addrspace(1) %out) #1 {
; SDAG-REAL16-LABEL: tanh_undef_bf16:
; SDAG-REAL16: ; %bb.0:
; SDAG-REAL16-NEXT: s_endpgm
;
; SDAG-FAKE16-LABEL: tanh_undef_bf16:
; SDAG-FAKE16: ; %bb.0:
; SDAG-FAKE16-NEXT: s_endpgm
%tanh = call bfloat @llvm.amdgcn.tanh.bf16(bfloat undef)
store bfloat %tanh, ptr addrspace(1) %out, align 2
ret void
}
attributes #0 = { nounwind readnone }
attributes #1 = { nounwind }
Reference in New Issue View Git Blame Copy Permalink