1 //===-- VOP3PInstructions.td - Vector Instruction Defintions --------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 //===----------------------------------------------------------------------===//
12 //===----------------------------------------------------------------------===//
14 class VOP3PInst<string OpName, VOPProfile P, SDPatternOperator node = null_frag> :
15 VOP3P_Pseudo<OpName, P,
16 !if(P.HasModifiers, getVOP3PModPat<P, node>.ret, getVOP3Pat<P, node>.ret)
19 // Non-packed instructions that use the VOP3P encoding.
20 // VOP3 neg/abs and VOP3P opsel/opsel_hi modifiers are allowed.
21 class VOP3_VOP3PInst<string OpName, VOPProfile P, SDPatternOperator node = null_frag> :
22 VOP3P_Pseudo<OpName, P> {
25 FP32InputMods:$src0_modifiers, VCSrc_f32:$src0,
26 FP32InputMods:$src1_modifiers, VCSrc_f32:$src1,
27 FP32InputMods:$src2_modifiers, VCSrc_f32:$src2,
30 op_sel_hi:$op_sel_hi);
32 " $vdst, $src0_modifiers, $src1_modifiers, $src2_modifiers$op_sel$op_sel_hi$clamp";
35 let isCommutable = 1 in {
36 def V_PK_FMA_F16 : VOP3PInst<"v_pk_fma_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16_V2F16>, fma>;
37 def V_PK_MAD_I16 : VOP3PInst<"v_pk_mad_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16_V2I16>>;
38 def V_PK_MAD_U16 : VOP3PInst<"v_pk_mad_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16_V2I16>>;
40 def V_PK_ADD_F16 : VOP3PInst<"v_pk_add_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16>, fadd>;
41 def V_PK_MUL_F16 : VOP3PInst<"v_pk_mul_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16>, fmul>;
42 def V_PK_MAX_F16 : VOP3PInst<"v_pk_max_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16>, fmaxnum>;
43 def V_PK_MIN_F16 : VOP3PInst<"v_pk_min_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16>, fminnum>;
45 def V_PK_ADD_U16 : VOP3PInst<"v_pk_add_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, add>;
46 def V_PK_ADD_I16 : VOP3PInst<"v_pk_add_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>>;
47 def V_PK_MUL_LO_U16 : VOP3PInst<"v_pk_mul_lo_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, mul>;
49 def V_PK_MIN_I16 : VOP3PInst<"v_pk_min_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, smin>;
50 def V_PK_MIN_U16 : VOP3PInst<"v_pk_min_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, umin>;
51 def V_PK_MAX_I16 : VOP3PInst<"v_pk_max_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, smax>;
52 def V_PK_MAX_U16 : VOP3PInst<"v_pk_max_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, umax>;
55 def V_PK_SUB_U16 : VOP3PInst<"v_pk_sub_u16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>>;
56 def V_PK_SUB_I16 : VOP3PInst<"v_pk_sub_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, sub>;
58 def V_PK_LSHLREV_B16 : VOP3PInst<"v_pk_lshlrev_b16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, lshl_rev>;
59 def V_PK_ASHRREV_I16 : VOP3PInst<"v_pk_ashrrev_i16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, ashr_rev>;
60 def V_PK_LSHRREV_B16 : VOP3PInst<"v_pk_lshrrev_b16", VOP3_Profile<VOP_V2I16_V2I16_V2I16>, lshr_rev>;
63 // These are VOP3a-like opcodes which accept no omod.
64 // Size of src arguments (16/32) is controlled by op_sel.
65 // For 16-bit src arguments their location (hi/lo) are controlled by op_sel_hi.
66 def V_MAD_MIX_F32 : VOP3_VOP3PInst<"v_mad_mix_f32", VOP3_Profile<VOP_F32_V2F16_V2F16_V2F16>>;
67 def V_MAD_MIXLO_F16 : VOP3_VOP3PInst<"v_mad_mixlo_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16_V2F16>>;
68 def V_MAD_MIXHI_F16 : VOP3_VOP3PInst<"v_mad_mixhi_f16", VOP3_Profile<VOP_V2F16_V2F16_V2F16_V2F16>>;
71 multiclass VOP3P_Real_vi<bits<10> op> {
72 def _vi : VOP3P_Real<!cast<VOP3P_Pseudo>(NAME), SIEncodingFamily.VI>,
73 VOP3Pe <op, !cast<VOP3P_Pseudo>(NAME).Pfl> {
74 let AssemblerPredicates = [HasVOP3PInsts];
75 let DecoderNamespace = "VI";
79 defm V_PK_MAD_I16 : VOP3P_Real_vi <0x380>;
80 defm V_PK_MUL_LO_U16 : VOP3P_Real_vi <0x381>;
81 defm V_PK_ADD_I16 : VOP3P_Real_vi <0x382>;
82 defm V_PK_SUB_I16 : VOP3P_Real_vi <0x383>;
83 defm V_PK_LSHLREV_B16 : VOP3P_Real_vi <0x384>;
84 defm V_PK_LSHRREV_B16 : VOP3P_Real_vi <0x385>;
85 defm V_PK_ASHRREV_I16 : VOP3P_Real_vi <0x386>;
86 defm V_PK_MAX_I16 : VOP3P_Real_vi <0x387>;
87 defm V_PK_MIN_I16 : VOP3P_Real_vi <0x388>;
88 defm V_PK_MAD_U16 : VOP3P_Real_vi <0x389>;
90 defm V_PK_ADD_U16 : VOP3P_Real_vi <0x38a>;
91 defm V_PK_SUB_U16 : VOP3P_Real_vi <0x38b>;
92 defm V_PK_MAX_U16 : VOP3P_Real_vi <0x38c>;
93 defm V_PK_MIN_U16 : VOP3P_Real_vi <0x38d>;
94 defm V_PK_FMA_F16 : VOP3P_Real_vi <0x38e>;
95 defm V_PK_ADD_F16 : VOP3P_Real_vi <0x38f>;
96 defm V_PK_MUL_F16 : VOP3P_Real_vi <0x390>;
97 defm V_PK_MIN_F16 : VOP3P_Real_vi <0x391>;
98 defm V_PK_MAX_F16 : VOP3P_Real_vi <0x392>;
100 defm V_MAD_MIX_F32 : VOP3P_Real_vi <0x3a0>;
101 defm V_MAD_MIXLO_F16 : VOP3P_Real_vi <0x3a1>;
102 defm V_MAD_MIXHI_F16 : VOP3P_Real_vi <0x3a2>;