d226f1b161
Regenerate checks after two recent commits that caused extra stuff to be added at the end of assembly lines, so the existing checks did not fail. - #179414 added "nv" to loads and stores on GFX1250. - #185774 added "msbs" comments on setreg instructions.
103 lines
4.6 KiB
LLVM
103 lines
4.6 KiB
LLVM
; 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 bfloat @llvm.amdgcn.rsq.bf16(bfloat) #0
|
|
|
|
define amdgpu_kernel void @rsq_bf16(ptr addrspace(1) %out, bfloat %src) #1 {
|
|
; SDAG-REAL16-LABEL: rsq_bf16:
|
|
; SDAG-REAL16: ; %bb.0:
|
|
; SDAG-REAL16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; SDAG-REAL16-NEXT: s_load_b96 s[0:2], s[4:5], 0x0 nv
|
|
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
|
|
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
|
|
; SDAG-REAL16-NEXT: v_rsq_bf16_e32 v0.l, s2
|
|
; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; SDAG-REAL16-NEXT: s_endpgm
|
|
;
|
|
; SDAG-FAKE16-LABEL: rsq_bf16:
|
|
; SDAG-FAKE16: ; %bb.0:
|
|
; SDAG-FAKE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; SDAG-FAKE16-NEXT: s_load_b96 s[0:2], s[4:5], 0x0 nv
|
|
; SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0
|
|
; SDAG-FAKE16-NEXT: s_wait_kmcnt 0x0
|
|
; SDAG-FAKE16-NEXT: v_rsq_bf16_e32 v0, s2
|
|
; SDAG-FAKE16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; SDAG-FAKE16-NEXT: s_endpgm
|
|
%rsq = call bfloat @llvm.amdgcn.rsq.bf16(bfloat %src) #0
|
|
store bfloat %rsq, ptr addrspace(1) %out, align 2
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @rsq_bf16_constant_4(ptr addrspace(1) %out) #1 {
|
|
; SDAG-REAL16-LABEL: rsq_bf16_constant_4:
|
|
; SDAG-REAL16: ; %bb.0:
|
|
; SDAG-REAL16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; SDAG-REAL16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 nv
|
|
; SDAG-REAL16-NEXT: v_rsq_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: global_store_b16 v1, v0, s[0:1]
|
|
; SDAG-REAL16-NEXT: s_endpgm
|
|
;
|
|
; SDAG-FAKE16-LABEL: rsq_bf16_constant_4:
|
|
; SDAG-FAKE16: ; %bb.0:
|
|
; SDAG-FAKE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; SDAG-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 nv
|
|
; SDAG-FAKE16-NEXT: v_rsq_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
|
|
%rsq = call bfloat @llvm.amdgcn.rsq.bf16(bfloat 4.0) #0
|
|
store bfloat %rsq, ptr addrspace(1) %out, align 2
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @rsq_bf16_constant_100(ptr addrspace(1) %out) #1 {
|
|
; SDAG-REAL16-LABEL: rsq_bf16_constant_100:
|
|
; SDAG-REAL16: ; %bb.0:
|
|
; SDAG-REAL16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; SDAG-REAL16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 nv
|
|
; SDAG-REAL16-NEXT: v_rsq_bf16_e32 v0.l, 0x42c8
|
|
; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0
|
|
; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0
|
|
; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1]
|
|
; SDAG-REAL16-NEXT: s_endpgm
|
|
;
|
|
; SDAG-FAKE16-LABEL: rsq_bf16_constant_100:
|
|
; SDAG-FAKE16: ; %bb.0:
|
|
; SDAG-FAKE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; SDAG-FAKE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 nv
|
|
; SDAG-FAKE16-NEXT: v_rsq_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
|
|
%rsq = call bfloat @llvm.amdgcn.rsq.bf16(bfloat 100.0) #0
|
|
store bfloat %rsq, ptr addrspace(1) %out, align 2
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @rsq_undef_bf16(ptr addrspace(1) %out) #1 {
|
|
; SDAG-REAL16-LABEL: rsq_undef_bf16:
|
|
; SDAG-REAL16: ; %bb.0:
|
|
; SDAG-REAL16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; SDAG-REAL16-NEXT: s_endpgm
|
|
;
|
|
; SDAG-FAKE16-LABEL: rsq_undef_bf16:
|
|
; SDAG-FAKE16: ; %bb.0:
|
|
; SDAG-FAKE16-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_WAVE_MODE, 25, 1), 1 ; msbs: dst=0 src0=0 src1=0 src2=0
|
|
; SDAG-FAKE16-NEXT: s_endpgm
|
|
%rsq = call bfloat @llvm.amdgcn.rsq.bf16(bfloat undef)
|
|
store bfloat %rsq, ptr addrspace(1) %out, align 2
|
|
ret void
|
|
}
|
|
|
|
attributes #0 = { nounwind readnone }
|
|
attributes #1 = { nounwind }
|