"Dummy feature to disable assembler instructions"
>;
-def FeatureGCN : SubtargetFeature<"gcn",
- "IsGCN",
- "true",
- "GCN or newer GPU"
->;
-
class GCNSubtargetFeatureGeneration <string Value,
list<SubtargetFeature> Implies> :
SubtargetFeatureGeneration <Value, "GCNSubtarget", Implies>;
def FeatureSouthernIslands : GCNSubtargetFeatureGeneration<"SOUTHERN_ISLANDS",
[FeatureFP64, FeatureLocalMemorySize32768, FeatureMIMG_R128,
- FeatureWavefrontSize64, FeatureGCN,
+ FeatureWavefrontSize64,
FeatureLDSBankCount32, FeatureMovrel, FeatureTrigReducedRange]
>;
def FeatureSeaIslands : GCNSubtargetFeatureGeneration<"SEA_ISLANDS",
[FeatureFP64, FeatureLocalMemorySize65536, FeatureMIMG_R128,
- FeatureWavefrontSize64, FeatureGCN, FeatureFlatAddressSpace,
+ FeatureWavefrontSize64, FeatureFlatAddressSpace,
FeatureCIInsts, FeatureMovrel, FeatureTrigReducedRange]
>;
def FeatureVolcanicIslands : GCNSubtargetFeatureGeneration<"VOLCANIC_ISLANDS",
[FeatureFP64, FeatureLocalMemorySize65536, FeatureMIMG_R128,
- FeatureWavefrontSize64, FeatureFlatAddressSpace, FeatureGCN,
+ FeatureWavefrontSize64, FeatureFlatAddressSpace,
FeatureGCN3Encoding, FeatureCIInsts, FeatureVIInsts, Feature16BitInsts,
FeatureSMemRealTime, FeatureVGPRIndexMode, FeatureMovrel,
FeatureScalarStores, FeatureInv2PiInlineImm,
def FeatureGFX9 : GCNSubtargetFeatureGeneration<"GFX9",
[FeatureFP64, FeatureLocalMemorySize65536,
- FeatureWavefrontSize64, FeatureFlatAddressSpace, FeatureGCN,
+ FeatureWavefrontSize64, FeatureFlatAddressSpace,
FeatureGCN3Encoding, FeatureCIInsts, FeatureVIInsts, Feature16BitInsts,
FeatureSMemRealTime, FeatureScalarStores, FeatureInv2PiInlineImm,
FeatureApertureRegs, FeatureGFX9Insts, FeatureVOP3P, FeatureVGPRIndexMode,
def TruePredicate : Predicate<"true">;
-// Exists to help track down where SubtargetPredicate isn't set rather
-// than letting tablegen crash with an unhelpful error.
-def InvalidPred : Predicate<"predicate not set on instruction or pattern">;
-
class PredicateControl {
- Predicate SubtargetPredicate = InvalidPred;
+ Predicate SubtargetPredicate = TruePredicate;
list<Predicate> AssemblerPredicates = [];
Predicate AssemblerPredicate = TruePredicate;
list<Predicate> OtherPredicates = [];
: AMDGPUPat<
(sub_type (extractelt vec_type:$src, sub_idx)),
(EXTRACT_SUBREG $src, sub_reg)
-> {
- let SubtargetPredicate = TruePredicate;
-}
+>;
/* Insert element pattern */
class Insert_Element <ValueType elem_type, ValueType vec_type,
: AMDGPUPat <
(insertelt vec_type:$vec, elem_type:$elem, sub_idx),
(INSERT_SUBREG $vec, $elem, sub_reg)
-> {
- let SubtargetPredicate = TruePredicate;
-}
+>;
// XXX: Convert to new syntax and use COPY_TO_REG, once the DFAPacketizer
// can handle COPY instructions.
HasMovrel = true;
}
+ // Don't crash on invalid devices.
+ if (WavefrontSize == 0)
+ WavefrontSize = 64;
+
HasFminFmaxLegacy = getGeneration() < AMDGPUSubtarget::VOLCANIC_ISLANDS;
return *this;
InstSI <outs, ins, "", pattern>,
SIMCInstr <opName, SIEncodingFamily.NONE> {
- let SubtargetPredicate = isGCN;
-
let LGKM_CNT = 1;
let DS = 1;
let Size = 8;
// The code produced for "generic" is only useful for tests and cannot
// reasonably be expected to execute on any particular target.
def : ProcessorModel<"generic", NoSchedModel,
- [FeatureGCN, FeatureWavefrontSize64]
+ [FeatureWavefrontSize64]
>;
//===----------------------------------------------------------------------===//
let UseNamedOperandTable = 1;
let hasSideEffects = 0; // XXX ????
- let SubtargetPredicate = isGCN;
let DecoderNamespace = dns;
let isAsmParserOnly = !if(!eq(dns,""), 1, 0);
let AsmMatchConverter = "cvtMIMG";
//
//===----------------------------------------------------------------------===//
-def isGCN : Predicate<"Subtarget->getGeneration() "
- ">= AMDGPUSubtarget::SOUTHERN_ISLANDS">,
- AssemblerPredicate<"FeatureGCN">;
def isSI : Predicate<"Subtarget->getGeneration() "
"== AMDGPUSubtarget::SOUTHERN_ISLANDS">,
AssemblerPredicate<"FeatureSouthernIslands">;
class InstSI <dag outs, dag ins, string asm = "",
list<dag> pattern = []> :
AMDGPUInst<outs, ins, asm, pattern>, GCNPredicateControl {
- let SubtargetPredicate = isGCN;
-
// Low bits - basic encoding information.
field bit SALU = 0;
field bit VALU = 0;
//===----------------------------------------------------------------------===//
class GCNPat<dag pattern, dag result> : Pat<pattern, result>, GCNPredicateControl {
- let SubtargetPredicate = isGCN;
+
}
include "SOPInstructions.td"
// VOP1 Patterns
//===----------------------------------------------------------------------===//
-let SubtargetPredicate = isGCN, OtherPredicates = [UnsafeFPMath] in {
+let OtherPredicates = [UnsafeFPMath] in {
//def : RcpPat<V_RCP_F64_e32, f64>;
//defm : RsqPat<V_RSQ_F64_e32, f64>;
(V_FRACT_F64_e64 $mods, $x, DSTCLAMP.NONE, DSTOMOD.NONE)
>;
-} // End SubtargetPredicate = isGCN, OtherPredicates = [UnsafeFPMath]
+} // End OtherPredicates = [UnsafeFPMath]
// f16_to_fp patterns
(v2f16 (EXTRACT_SUBREG v4f16:$vec, sub1))
>;
-let SubtargetPredicate = isGCN in {
-
// FIXME: Why do only some of these type combinations for SReg and
// VReg?
// 16-bit bitcast
def : BitConvert <v16i32, v16f32, VReg_512>;
def : BitConvert <v16f32, v16i32, VReg_512>;
-} // End SubtargetPredicate = isGCN
-
/********** =================== **********/
/********** Src & Dst modifiers **********/
/********** =================== **********/
/********** Intrinsic Patterns **********/
/********** ================== **********/
-let SubtargetPredicate = isGCN in {
def : POW_Common <V_LOG_F32_e32, V_EXP_F32_e32, V_MUL_LEGACY_F32_e32>;
-}
def : GCNPat <
(i32 (sext i1:$src0)),
// VOP3 Patterns
//===----------------------------------------------------------------------===//
-let SubtargetPredicate = isGCN in {
-
def : IMad24Pat<V_MAD_I32_I24, 1>;
def : UMad24Pat<V_MAD_U32_U24, 1>;
defm : BFIPatterns <V_BFI_B32, S_MOV_B32, SReg_64>;
def : ROTRPattern <V_ALIGNBIT_B32>;
-}
-
def : GCNPat<(i32 (trunc (srl i64:$src0, (and i32:$src1, (i32 31))))),
(V_ALIGNBIT_B32 (i32 (EXTRACT_SUBREG (i64 $src0), sub1)),
(i32 (EXTRACT_SUBREG (i64 $src0), sub0)), $src1)>;
>;
}
-let SubtargetPredicate = isGCN in {
-
defm : BFMPatterns <i32, S_BFM_B32, S_MOV_B32>;
// FIXME: defm : BFMPatterns <i64, S_BFM_B64, S_MOV_B64>;
defm : IntMed3Pat<V_MED3_I32, smin, smax, smin_oneuse, smax_oneuse>;
defm : IntMed3Pat<V_MED3_U32, umin, umax, umin_oneuse, umax_oneuse>;
-}
-
// This matches 16 permutations of
// max(min(x, y), min(max(x, y), z))
class FPMed3Pat<ValueType vt,
let hasSideEffects = 0;
let UseNamedOperandTable = 1;
let SchedRW = [WriteSMEM];
- let SubtargetPredicate = isGCN;
string Mnemonic = opName;
string AsmOperands = asmOps;
let isPseudo = 1;
let isCodeGenOnly = 1;
- let SubtargetPredicate = isGCN;
string Mnemonic = opName;
string AsmOperands = asmOps;
SIMCInstr<opName, SIEncodingFamily.NONE> {
let isPseudo = 1;
let isCodeGenOnly = 1;
- let SubtargetPredicate = isGCN;
let mayLoad = 0;
let mayStore = 0;
let hasSideEffects = 0;
let Defs = [SCC];
let SchedRW = [WriteSALU];
let UseNamedOperandTable = 1;
- let SubtargetPredicate = isGCN;
}
class SOPC_Base <bits<7> op, RegisterOperand rc0, RegisterOperand rc1,
let SchedRW = [WriteSALU];
let UseNamedOperandTable = 1;
- let SubtargetPredicate = isGCN;
}
unsigned getMaxWorkGroupsPerCU(const MCSubtargetInfo *STI,
unsigned FlatWorkGroupSize) {
- if (!STI->getFeatureBits().test(FeatureGCN))
+ assert(FlatWorkGroupSize != 0);
+ if (STI->getTargetTriple().getArch() != Triple::amdgcn)
return 8;
unsigned N = getWavesPerWorkGroup(STI, FlatWorkGroupSize);
if (N == 1)
let mayLoad = 0;
let mayStore = 0;
let hasSideEffects = 0;
- let SubtargetPredicate = isGCN;
let VOP1 = 1;
let VALU = 1;
let mayLoad = 0;
let mayStore = 0;
let hasSideEffects = 0;
- let SubtargetPredicate = isGCN;
let VOP1 = 1;
let VALU = 1;
let mayLoad = 0;
let mayStore = 0;
let hasSideEffects = 0;
- let SubtargetPredicate = isGCN;
let VOP2 = 1;
let VALU = 1;
// VOP2 Instructions
//===----------------------------------------------------------------------===//
-let SubtargetPredicate = isGCN, Predicates = [isGCN] in {
-
defm V_CNDMASK_B32 : VOP2eInst <"v_cndmask_b32", VOP2e_I32_I32_I32_I1>;
def V_MADMK_F32 : VOP2_Pseudo <"v_madmk_f32", VOP_MADMK_F32, []>;
defm V_CVT_PK_U16_U32 : VOP2Inst <"v_cvt_pk_u16_u32", VOP_NO_EXT<VOP_V2I16_I32_I32>, AMDGPUpk_u16_u32>;
defm V_CVT_PK_I16_I32 : VOP2Inst <"v_cvt_pk_i16_i32", VOP_NO_EXT<VOP_V2I16_I32_I32>, AMDGPUpk_i16_i32>;
-} // End SubtargetPredicate = isGCN, Predicates = [isGCN]
def : GCNPat<
(AMDGPUadde i32:$src0, i32:$src1, i1:$src2),
let Uses = [EXEC];
let Defs = [VCC];
- let SubtargetPredicate = isGCN;
-
VOPProfile Pfl = P;
}
let mayLoad = 0;
let mayStore = 0;
let hasSideEffects = 0;
- let SubtargetPredicate = isGCN;
// Because SGPRs may be allowed if there are multiple operands, we
// need a post-isel hook to insert copies in order to avoid
; RUN: llc -mtriple=amdgcn--mesa3d -mcpu=gfx600 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; RUN: llc -mtriple=amdgcn--mesa3d -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
-; RUN: llc -mtriple=amdgcn--mesa3d -mcpu=gfx800 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
+; RUN: llc -mtriple=amdgcn--mesa3d -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; RUN: llc -mtriple=amdgcn--mesa3d -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; FUNC-LABEL: {{^}}system_acquire: