aboutsummaryrefslogtreecommitdiff
path: root/lib/Target/AMDGPU/VOP3Instructions.td
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Target/AMDGPU/VOP3Instructions.td')
-rw-r--r--lib/Target/AMDGPU/VOP3Instructions.td57
1 files changed, 30 insertions, 27 deletions
diff --git a/lib/Target/AMDGPU/VOP3Instructions.td b/lib/Target/AMDGPU/VOP3Instructions.td
index aedbfa015bf6..17ae08dc6267 100644
--- a/lib/Target/AMDGPU/VOP3Instructions.td
+++ b/lib/Target/AMDGPU/VOP3Instructions.td
@@ -153,19 +153,24 @@ class getVOP3VCC<VOPProfile P, SDPatternOperator node> {
(i1 VCC)))];
}
-class VOP3Features<bit Clamp, bit OpSel> {
+class VOP3Features<bit Clamp, bit OpSel, bit Packed> {
bit HasClamp = Clamp;
bit HasOpSel = OpSel;
+ bit IsPacked = Packed;
}
-def VOP3_REGULAR : VOP3Features<0, 0>;
-def VOP3_CLAMP : VOP3Features<1, 0>;
-def VOP3_OPSEL : VOP3Features<1, 1>;
+def VOP3_REGULAR : VOP3Features<0, 0, 0>;
+def VOP3_CLAMP : VOP3Features<1, 0, 0>;
+def VOP3_OPSEL : VOP3Features<1, 1, 0>;
+def VOP3_PACKED : VOP3Features<1, 1, 1>;
class VOP3_Profile<VOPProfile P, VOP3Features Features = VOP3_REGULAR> : VOPProfile<P.ArgVT> {
let HasClamp = !if(Features.HasClamp, 1, P.HasClamp);
let HasOpSel = !if(Features.HasOpSel, 1, P.HasOpSel);
+ let IsPacked = !if(Features.IsPacked, 1, P.IsPacked);
+
+ let HasModifiers = !if(Features.IsPacked, 1, P.HasModifiers);
// FIXME: Hack to stop printing _e64
let Outs64 = (outs DstRC.RegClass:$vdst);
@@ -283,10 +288,10 @@ def V_MAD_F32 : VOP3Inst <"v_mad_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, fmad>;
def V_MAD_I32_I24 : VOP3Inst <"v_mad_i32_i24", VOP3_Profile<VOP_I32_I32_I32_I32, VOP3_CLAMP>>;
def V_MAD_U32_U24 : VOP3Inst <"v_mad_u32_u24", VOP3_Profile<VOP_I32_I32_I32_I32, VOP3_CLAMP>>;
def V_FMA_F32 : VOP3Inst <"v_fma_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, fma>;
-def V_FMA_F64 : VOP3Inst <"v_fma_f64", VOP3_Profile<VOP_F64_F64_F64_F64>, fma>;
def V_LERP_U8 : VOP3Inst <"v_lerp_u8", VOP3_Profile<VOP_I32_I32_I32_I32>, int_amdgcn_lerp>;
let SchedRW = [WriteDoubleAdd] in {
+def V_FMA_F64 : VOP3Inst <"v_fma_f64", VOP3_Profile<VOP_F64_F64_F64_F64>, fma>;
def V_ADD_F64 : VOP3Inst <"v_add_f64", VOP3_Profile<VOP_F64_F64_F64>, fadd, 1>;
def V_MUL_F64 : VOP3Inst <"v_mul_f64", VOP3_Profile<VOP_F64_F64_F64>, fmul, 1>;
def V_MIN_F64 : VOP3Inst <"v_min_f64", VOP3_Profile<VOP_F64_F64_F64>, fminnum, 1>;
@@ -355,14 +360,12 @@ def V_LDEXP_F64 : VOP3Inst <"v_ldexp_f64", VOP3_Profile<VOP_F64_F64_I32>, AMDGPU
def V_DIV_SCALE_F32 : VOP3_Pseudo <"v_div_scale_f32", VOP3b_F32_I1_F32_F32_F32, [], 1> {
let SchedRW = [WriteFloatFMA, WriteSALU];
- let hasExtraSrcRegAllocReq = 1;
let AsmMatchConverter = "";
}
// Double precision division pre-scale.
def V_DIV_SCALE_F64 : VOP3_Pseudo <"v_div_scale_f64", VOP3b_F64_I1_F64_F64_F64, [], 1> {
let SchedRW = [WriteDouble, WriteSALU];
- let hasExtraSrcRegAllocReq = 1;
let AsmMatchConverter = "";
}
@@ -376,6 +379,7 @@ def V_TRIG_PREOP_F64 : VOP3Inst <"v_trig_preop_f64", VOP3_Profile<VOP_F64_F64_I3
let SchedRW = [WriteDouble];
}
+let SchedRW = [Write64Bit] in {
// These instructions only exist on SI and CI
let SubtargetPredicate = isSICI in {
def V_LSHL_B64 : VOP3Inst <"v_lshl_b64", VOP3_Profile<VOP_I64_I64_I32>>;
@@ -389,17 +393,17 @@ def V_LSHLREV_B64 : VOP3Inst <"v_lshlrev_b64", VOP3_Profile<VOP_I64_I32_I64>>;
def V_LSHRREV_B64 : VOP3Inst <"v_lshrrev_b64", VOP3_Profile<VOP_I64_I32_I64>>;
def V_ASHRREV_I64 : VOP3Inst <"v_ashrrev_i64", VOP3_Profile<VOP_I64_I32_I64>>;
} // End SubtargetPredicate = isVI
-
+} // End SchedRW = [Write64Bit]
let SubtargetPredicate = isCIVI in {
-let Constraints = "@earlyclobber $vdst" in {
+let Constraints = "@earlyclobber $vdst", SchedRW = [WriteQuarterRate32] in {
def V_QSAD_PK_U16_U8 : VOP3Inst <"v_qsad_pk_u16_u8", VOP3_Profile<VOP_I64_I64_I32_I64, VOP3_CLAMP>>;
def V_MQSAD_U32_U8 : VOP3Inst <"v_mqsad_u32_u8", VOP3_Profile<VOP_V4I32_I64_I32_V4I32, VOP3_CLAMP>>;
-} // End Constraints = "@earlyclobber $vdst"
+} // End Constraints = "@earlyclobber $vdst", SchedRW = [WriteQuarterRate32]
let isCommutable = 1 in {
-let SchedRW = [WriteDouble, WriteSALU] in {
+let SchedRW = [WriteQuarterRate32, WriteSALU] in {
def V_MAD_U64_U32 : VOP3Inst <"v_mad_u64_u32", VOP3b_I64_I1_I32_I32_I64>;
def V_MAD_I64_I32 : VOP3Inst <"v_mad_i64_i32", VOP3b_I64_I1_I32_I32_I64>;
} // End SchedRW = [WriteDouble, WriteSALU]
@@ -408,16 +412,16 @@ def V_MAD_I64_I32 : VOP3Inst <"v_mad_i64_i32", VOP3b_I64_I1_I32_I32_I64>;
} // End SubtargetPredicate = isCIVI
-let SubtargetPredicate = Has16BitInsts in {
-
-let renamedInGFX9 = 1 in {
-def V_DIV_FIXUP_F16 : VOP3Inst <"v_div_fixup_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, AMDGPUdiv_fixup>;
+def V_DIV_FIXUP_F16 : VOP3Inst <"v_div_fixup_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, AMDGPUdiv_fixup> {
+ let Predicates = [Has16BitInsts, isVIOnly];
}
-let SubtargetPredicate = isGFX9 in {
-def V_DIV_FIXUP_F16_gfx9 : VOP3Inst <"v_div_fixup_f16_gfx9", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>>;
+def V_DIV_FIXUP_F16_gfx9 : VOP3Inst <"v_div_fixup_f16_gfx9",
+ VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, AMDGPUdiv_fixup> {
+ let renamedInGFX9 = 1;
+ let Predicates = [Has16BitInsts, isGFX9];
}
-let isCommutable = 1 in {
+let SubtargetPredicate = Has16BitInsts, isCommutable = 1 in {
let renamedInGFX9 = 1 in {
def V_MAD_F16 : VOP3Inst <"v_mad_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, fmad>;
@@ -438,15 +442,14 @@ def V_INTERP_P2_F16_gfx9 : VOP3Interp <"v_interp_p2_f16_gfx9", VOP3_INTERP16<[f1
def V_INTERP_P1LL_F16 : VOP3Interp <"v_interp_p1ll_f16", VOP3_INTERP16<[f32, f32, i32, untyped]>>;
def V_INTERP_P1LV_F16 : VOP3Interp <"v_interp_p1lv_f16", VOP3_INTERP16<[f32, f32, i32, f16]>>;
-} // End isCommutable = 1
-} // End SubtargetPredicate = Has16BitInsts
+} // End SubtargetPredicate = Has16BitInsts, isCommutable = 1
let SubtargetPredicate = isVI in {
def V_INTERP_P1_F32_e64 : VOP3Interp <"v_interp_p1_f32", VOP3_INTERP>;
def V_INTERP_P2_F32_e64 : VOP3Interp <"v_interp_p2_f32", VOP3_INTERP>;
def V_INTERP_MOV_F32_e64 : VOP3Interp <"v_interp_mov_f32", VOP3_INTERP_MOV>;
-def V_PERM_B32 : VOP3Inst <"v_perm_b32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
+def V_PERM_B32 : VOP3Inst <"v_perm_b32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUperm>;
} // End SubtargetPredicate = isVI
let Predicates = [Has16BitInsts] in {
@@ -697,7 +700,7 @@ multiclass VOP3Interp_F16_Real_vi<bits<10> op> {
let AssemblerPredicates = [isGFX9], DecoderNamespace = "GFX9" in {
multiclass VOP3_F16_Real_gfx9<bits<10> op, string OpName, string AsmName> {
- def _vi : VOP3_Real<!cast<VOP3_Pseudo>(OpName), SIEncodingFamily.GFX9>,
+ def _gfx9 : VOP3_Real<!cast<VOP3_Pseudo>(OpName), SIEncodingFamily.GFX9>,
VOP3e_vi <op, !cast<VOP3_Pseudo>(OpName).Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(OpName);
let AsmString = AsmName # ps.AsmOperands;
@@ -705,7 +708,7 @@ multiclass VOP3_F16_Real_gfx9<bits<10> op, string OpName, string AsmName> {
}
multiclass VOP3OpSel_F16_Real_gfx9<bits<10> op, string AsmName> {
- def _vi : VOP3_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.GFX9>,
+ def _gfx9 : VOP3_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.GFX9>,
VOP3OpSel_gfx9 <op, !cast<VOP3_Pseudo>(NAME).Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(NAME);
let AsmString = AsmName # ps.AsmOperands;
@@ -713,7 +716,7 @@ multiclass VOP3OpSel_F16_Real_gfx9<bits<10> op, string AsmName> {
}
multiclass VOP3Interp_F16_Real_gfx9<bits<10> op, string OpName, string AsmName> {
- def _vi : VOP3_Real<!cast<VOP3_Pseudo>(OpName), SIEncodingFamily.GFX9>,
+ def _gfx9 : VOP3_Real<!cast<VOP3_Pseudo>(OpName), SIEncodingFamily.GFX9>,
VOP3Interp_vi <op, !cast<VOP3_Pseudo>(OpName).Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(OpName);
let AsmString = AsmName # ps.AsmOperands;
@@ -721,9 +724,9 @@ multiclass VOP3Interp_F16_Real_gfx9<bits<10> op, string OpName, string AsmName>
}
multiclass VOP3_Real_gfx9<bits<10> op, string AsmName> {
- def _vi : VOP3_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.GFX9>,
- VOP3e_vi <op, !cast<VOP3_Pseudo>(NAME).Pfl> {
- VOP3_Pseudo ps = !cast<VOP3_Pseudo>(NAME);
+ def _gfx9 : VOP3_Real<!cast<VOP_Pseudo>(NAME), SIEncodingFamily.GFX9>,
+ VOP3e_vi <op, !cast<VOP_Pseudo>(NAME).Pfl> {
+ VOP_Pseudo ps = !cast<VOP_Pseudo>(NAME);
let AsmString = AsmName # ps.AsmOperands;
}
}
generated by cgit v1.2.3 (git 2.25.1) at 2024-04-25 05:43:58 +0000