maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/fmuladd.f16.ll

468 lines
20 KiB
LLVM
Raw Normal View History

AMDGPU: Form more FMAs if fusion is allowed Extend the existing fadd/fsub->fmad combines to produce FMA if allowed. llvm-svn: 290311
2016-12-22 11:55:35 +08:00
; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-STRICT,VI-FLUSH,VI %s
; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-STRICT,VI-FLUSH,VI %s
; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-CONTRACT,VI-FLUSH,VI %s
; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-CONTRACT,VI-FLUSH,VI %s
; RUN: llc -march=amdgcn -mcpu=fiji -mattr=+fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-STRICT,VI-DENORM-STRICT,VI-DENORM,VI %s
; RUN: llc -march=amdgcn -mcpu=fiji -mattr=+fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-STRICT,VI-DENORM-STRICT,VI-DENORM,VI %s
; RUN: llc -march=amdgcn -mcpu=fiji -mattr=+fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-CONTRACT,VI-DENORM-CONTRACT,VI-DENORM,VI %s
; RUN: llc -march=amdgcn -mcpu=fiji -mattr=+fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-CONTRACT,VI-DENORM-CONTRACT,VI-DENORM,VI %s
declare i32 @llvm.amdgcn.workitem.id.x() #1
declare half @llvm.fmuladd.f16(half, half, half) #1
declare half @llvm.fabs.f16(half) #1
; GCN-LABEL: {{^}}fmuladd_f16:
; VI-FLUSH: v_mac_f16_e32 {{v[0-9]+, v[0-9]+, v[0-9]+}}
; VI-DENORM: v_fma_f16 {{v[0-9]+, v[0-9]+, v[0-9]+}}
define void @fmuladd_f16(half addrspace(1)* %out, half addrspace(1)* %in1,
half addrspace(1)* %in2, half addrspace(1)* %in3) #0 {
%r0 = load half, half addrspace(1)* %in1
%r1 = load half, half addrspace(1)* %in2
%r2 = load half, half addrspace(1)* %in3
%r3 = tail call half @llvm.fmuladd.f16(half %r0, half %r1, half %r2)
store half %r3, half addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}fmuladd_2.0_a_b_f16
; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
; VI-FLUSH: v_mac_f16_e32 [[R2]], 2.0, [[R1]]
; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @fmuladd_2.0_a_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
%r1 = load volatile half, half addrspace(1)* %gep.0
%r2 = load volatile half, half addrspace(1)* %gep.1
%r3 = tail call half @llvm.fmuladd.f16(half 2.0, half %r1, half %r2)
store half %r3, half addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}fmuladd_a_2.0_b_f16
; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
; VI-FLUSH: v_mac_f16_e32 [[R2]], 2.0, [[R1]]
; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @fmuladd_a_2.0_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
%r1 = load volatile half, half addrspace(1)* %gep.0
%r2 = load volatile half, half addrspace(1)* %gep.1
%r3 = tail call half @llvm.fmuladd.f16(half %r1, half 2.0, half %r2)
store half %r3, half addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}fadd_a_a_b_f16:
; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
; VI-FLUSH: v_mac_f16_e32 [[R2]], 2.0, [[R1]]
; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
; VI-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
; VI-DENORM-STRICT: v_add_f16_e32 [[RESULT:v[0-9]+]], [[R2]], [[TMP]]
; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @fadd_a_a_b_f16(half addrspace(1)* %out,
half addrspace(1)* %in1,
half addrspace(1)* %in2) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
%r0 = load volatile half, half addrspace(1)* %gep.0
%r1 = load volatile half, half addrspace(1)* %gep.1
%add.0 = fadd half %r0, %r0
%add.1 = fadd half %add.0, %r1
store half %add.1, half addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}fadd_b_a_a_f16:
; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
; VI-FLUSH: v_mac_f16_e32 [[R2]], 2.0, [[R1]]
; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
; VI-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
; VI-DENORM-STRICT: v_add_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[R2]]
; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @fadd_b_a_a_f16(half addrspace(1)* %out,
half addrspace(1)* %in1,
half addrspace(1)* %in2) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
%r0 = load volatile half, half addrspace(1)* %gep.0
%r1 = load volatile half, half addrspace(1)* %gep.1
%add.0 = fadd half %r0, %r0
%add.1 = fadd half %r1, %add.0
store half %add.1, half addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}fmuladd_neg_2.0_a_b_f16
; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
; VI-FLUSH: v_mac_f16_e32 [[R2]], -2.0, [[R1]]
; VI-DENORM: v_fma_f16 [[R2:v[0-9]+]], [[R1]], -2.0, [[R2]]
; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
define void @fmuladd_neg_2.0_a_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
%r1 = load volatile half, half addrspace(1)* %gep.0
%r2 = load volatile half, half addrspace(1)* %gep.1
%r3 = tail call half @llvm.fmuladd.f16(half -2.0, half %r1, half %r2)
store half %r3, half addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}fmuladd_neg_2.0_neg_a_b_f16
; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
; VI-FLUSH: v_mac_f16_e32 [[R2]], 2.0, [[R1]]
; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], -[[R1]], -2.0, [[R2]]
; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @fmuladd_neg_2.0_neg_a_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
%r1 = load volatile half, half addrspace(1)* %gep.0
%r2 = load volatile half, half addrspace(1)* %gep.1
%r1.fneg = fsub half -0.000000e+00, %r1
%r3 = tail call half @llvm.fmuladd.f16(half -2.0, half %r1.fneg, half %r2)
store half %r3, half addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}fmuladd_2.0_neg_a_b_f16
; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
; VI-FLUSH: v_mac_f16_e32 [[R2]], -2.0, [[R1]]
; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], -[[R1]], 2.0, [[R2]]
; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @fmuladd_2.0_neg_a_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
%r1 = load volatile half, half addrspace(1)* %gep.0
%r2 = load volatile half, half addrspace(1)* %gep.1
%r1.fneg = fsub half -0.000000e+00, %r1
%r3 = tail call half @llvm.fmuladd.f16(half 2.0, half %r1.fneg, half %r2)
store half %r3, half addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}fmuladd_2.0_a_neg_b_f16
; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
AMDGPU: Swap order of operands in fadd/fsub combine FMA is canonicalized to constant in the middle operand. Do the same so fmad matches and avoid an extra combine step. llvm-svn: 290313
2016-12-22 12:03:40 +08:00
; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, -[[R2]]
AMDGPU: Form more FMAs if fusion is allowed Extend the existing fadd/fsub->fmad combines to produce FMA if allowed. llvm-svn: 290311
2016-12-22 11:55:35 +08:00
; VI-DENORM: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, -[[R2]]
; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @fmuladd_2.0_a_neg_b_f16(half addrspace(1)* %out, half addrspace(1)* %in) #0 {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
%r1 = load volatile half, half addrspace(1)* %gep.0
%r2 = load volatile half, half addrspace(1)* %gep.1
%r2.fneg = fsub half -0.000000e+00, %r2
%r3 = tail call half @llvm.fmuladd.f16(half 2.0, half %r1, half %r2.fneg)
store half %r3, half addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}mad_sub_f16:
; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -[[REGC]]
; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -[[REGC]]
; VI-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGB]], [[REGA]]
; VI-DENORM-STRICT: v_subrev_f16_e32 [[RESULT:v[0-9]+]], [[REGC]], [[TMP]]
; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @mad_sub_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
%a = load volatile half, half addrspace(1)* %gep0, align 2
%b = load volatile half, half addrspace(1)* %gep1, align 2
%c = load volatile half, half addrspace(1)* %gep2, align 2
%mul = fmul half %a, %b
%sub = fsub half %mul, %c
store half %sub, half addrspace(1)* %outgep, align 2
ret void
}
; GCN-LABEL: {{^}}mad_sub_inv_f16:
; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], [[REGC]]
; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], [[REGC]]
; VI-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGB]], [[REGA]]
; VI-DENORM-STRICT: v_subrev_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[REGC]]
; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @mad_sub_inv_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
%a = load volatile half, half addrspace(1)* %gep0, align 2
%b = load volatile half, half addrspace(1)* %gep1, align 2
%c = load volatile half, half addrspace(1)* %gep2, align 2
%mul = fmul half %a, %b
%sub = fsub half %c, %mul
store half %sub, half addrspace(1)* %outgep, align 2
ret void
}
; GCN-LABEL: {{^}}mad_sub_fabs_f16:
; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -|[[REGC]]|
; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -|[[REGC]]|
; VI-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGB]], [[REGA]]
; VI-DENORM-STRICT: v_sub_f16_e64 [[RESULT:v[0-9]+]], [[TMP]], |[[REGC]]|
; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @mad_sub_fabs_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
%a = load volatile half, half addrspace(1)* %gep0, align 2
%b = load volatile half, half addrspace(1)* %gep1, align 2
%c = load volatile half, half addrspace(1)* %gep2, align 2
%c.abs = call half @llvm.fabs.f16(half %c) #0
%mul = fmul half %a, %b
%sub = fsub half %mul, %c.abs
store half %sub, half addrspace(1)* %outgep, align 2
ret void
}
; GCN-LABEL: {{^}}mad_sub_fabs_inv_f16:
; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], |[[REGC]]|
; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], |[[REGC]]|
; VI-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGB]], [[REGA]]
; VI-DENORM-STRICT: v_sub_f16_e64 [[RESULT:v[0-9]+]], |[[REGC]]|, [[TMP]]
; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @mad_sub_fabs_inv_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
%a = load volatile half, half addrspace(1)* %gep0, align 2
%b = load volatile half, half addrspace(1)* %gep1, align 2
%c = load volatile half, half addrspace(1)* %gep2, align 2
%c.abs = call half @llvm.fabs.f16(half %c) #0
%mul = fmul half %a, %b
%sub = fsub half %c.abs, %mul
store half %sub, half addrspace(1)* %outgep, align 2
ret void
}
; GCN-LABEL: {{^}}neg_neg_mad_f16:
; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
; VI-FLUSH: v_mac_f16_e32 [[REGC]], [[REGB]], [[REGA]]
; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REGC]]
; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], [[REGC]]
; VI-DENORM-STRICT: v_mul_f16_e32 [[TMP:v[0-9]+]], [[REGB]], [[REGA]]
; VI-DENORM-STRICT: v_add_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[REGC]]
; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @neg_neg_mad_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
%a = load volatile half, half addrspace(1)* %gep0, align 2
%b = load volatile half, half addrspace(1)* %gep1, align 2
%c = load volatile half, half addrspace(1)* %gep2, align 2
%nega = fsub half -0.000000e+00, %a
%negb = fsub half -0.000000e+00, %b
%mul = fmul half %nega, %negb
%sub = fadd half %mul, %c
store half %sub, half addrspace(1)* %outgep, align 2
ret void
}
; GCN-LABEL: {{^}}mad_fabs_sub_f16:
; GCN: {{buffer|flat}}_load_ushort [[REGA:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGB:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[REGC:v[0-9]+]]
; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], [[REGA]], |[[REGB]]|, -[[REGC]]
; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[REGA]], |[[REGB]]|, -[[REGC]]
; VI-DENORM-STRICT: v_mul_f16_e64 [[TMP:v[0-9]+]], [[REGA]], |[[REGB]]|
; VI-DENORM-STRICT: v_subrev_f16_e32 [[RESULT:v[0-9]+]], [[REGC]], [[TMP]]
; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @mad_fabs_sub_f16(half addrspace(1)* noalias nocapture %out, half addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr half, half addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr half, half addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr half, half addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr half, half addrspace(1)* %out, i64 %tid.ext
%a = load volatile half, half addrspace(1)* %gep0, align 2
%b = load volatile half, half addrspace(1)* %gep1, align 2
%c = load volatile half, half addrspace(1)* %gep2, align 2
%b.abs = call half @llvm.fabs.f16(half %b) #0
%mul = fmul half %a, %b.abs
%sub = fsub half %mul, %c
store half %sub, half addrspace(1)* %outgep, align 2
ret void
}
; GCN-LABEL: {{^}}fsub_c_fadd_a_a_f16:
; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
; VI-FLUSH: v_mac_f16_e32 [[R2]], -2.0, [[R1]]
; VI-FLUSH: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[R2]]
; VI-DENORM-CONTRACT: v_fma_f16 [[RESULT:v[0-9]+]], [[R1]], -2.0, [[R2]]
; VI-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
; VI-DENORM-STRICT: v_subrev_f16_e32 [[RESULT:v[0-9]+]], [[TMP]], [[R2]]
; VI-DENORM: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @fsub_c_fadd_a_a_f16(half addrspace(1)* %out, half addrspace(1)* %in) {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
%r1 = load volatile half, half addrspace(1)* %gep.0
%r2 = load volatile half, half addrspace(1)* %gep.1
%add = fadd half %r1, %r1
%r3 = fsub half %r2, %add
store half %r3, half addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}fsub_fadd_a_a_c_f16:
; GCN: {{buffer|flat}}_load_ushort [[R1:v[0-9]+]],
; GCN: {{buffer|flat}}_load_ushort [[R2:v[0-9]+]],
AMDGPU: Swap order of operands in fadd/fsub combine FMA is canonicalized to constant in the middle operand. Do the same so fmad matches and avoid an extra combine step. llvm-svn: 290313
2016-12-22 12:03:40 +08:00
; VI-FLUSH: v_mad_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, -[[R2]]
AMDGPU: Form more FMAs if fusion is allowed Extend the existing fadd/fsub->fmad combines to produce FMA if allowed. llvm-svn: 290311
2016-12-22 11:55:35 +08:00
; VI-DENORM-CONTRACT: v_fma_f16 [[R2]], [[R1]], 2.0, -[[R2]]
; VI-DENORM-STRICT: v_add_f16_e32 [[TMP:v[0-9]+]], [[R1]], [[R1]]
; VI-DENORM-STRICT: v_subrev_f16_e32 [[RESULT:v[0-9]+]], [[R2]], [[TMP]]
; VI: flat_store_short v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
define void @fsub_fadd_a_a_c_f16(half addrspace(1)* %out, half addrspace(1)* %in) {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep.0 = getelementptr half, half addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr half, half addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr half, half addrspace(1)* %out, i32 %tid
%r1 = load volatile half, half addrspace(1)* %gep.0
%r2 = load volatile half, half addrspace(1)* %gep.1
%add = fadd half %r1, %r1
%r3 = fsub half %add, %r2
store half %r3, half addrspace(1)* %gep.out
ret void
}
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }