diff --git a/lib/Target/R600/R600InstrFormats.td b/lib/Target/R600/R600InstrFormats.td index 2d72404702f..514ad4a7c5f 100644 --- a/lib/Target/R600/R600InstrFormats.td +++ b/lib/Target/R600/R600InstrFormats.td @@ -16,7 +16,6 @@ class InstR600 pattern, : AMDGPUInst { field bits<64> Inst; - bit TransOnly = 0; bit Trig = 0; bit Op3 = 0; bit isVector = 0; @@ -37,7 +36,6 @@ class InstR600 pattern, let Pattern = pattern; let Itinerary = itin; - let TSFlags{0} = TransOnly; let TSFlags{4} = Trig; let TSFlags{5} = Op3; diff --git a/lib/Target/R600/R600InstrInfo.cpp b/lib/Target/R600/R600InstrInfo.cpp index 3bc170f538b..83f8e71fbfa 100644 --- a/lib/Target/R600/R600InstrInfo.cpp +++ b/lib/Target/R600/R600InstrInfo.cpp @@ -150,13 +150,23 @@ bool R600InstrInfo::isLDSInstr(unsigned Opcode) const { } bool R600InstrInfo::isTransOnly(unsigned Opcode) const { - return (get(Opcode).TSFlags & R600_InstFlag::TRANS_ONLY); + if (ST.hasCaymanISA()) + return false; + return (get(Opcode).getSchedClass() == AMDGPU::TransALU); } bool R600InstrInfo::isTransOnly(const MachineInstr *MI) const { return isTransOnly(MI->getOpcode()); } +bool R600InstrInfo::isVectorOnly(unsigned Opcode) const { + return (get(Opcode).getSchedClass() == AMDGPU::VecALU); +} + +bool R600InstrInfo::isVectorOnly(const MachineInstr *MI) const { + return isVectorOnly(MI->getOpcode()); +} + bool R600InstrInfo::usesVertexCache(unsigned Opcode) const { return ST.hasVertexCache() && IS_VTX(get(Opcode)); } diff --git a/lib/Target/R600/R600InstrInfo.h b/lib/Target/R600/R600InstrInfo.h index cdaa2fbefc8..168306a1693 100644 --- a/lib/Target/R600/R600InstrInfo.h +++ b/lib/Target/R600/R600InstrInfo.h @@ -68,6 +68,8 @@ namespace llvm { bool isTransOnly(unsigned Opcode) const; bool isTransOnly(const MachineInstr *MI) const; + bool isVectorOnly(unsigned Opcode) const; + bool isVectorOnly(const MachineInstr *MI) const; bool usesVertexCache(unsigned Opcode) const; bool usesVertexCache(const MachineInstr *MI) const; diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td index 178e081e9db..48b055381a3 100644 --- a/lib/Target/R600/R600Instructions.td +++ b/lib/Target/R600/R600Instructions.td @@ -909,12 +909,16 @@ class CNDE_Common inst> : R600_3OP < class CNDGT_Common inst> : R600_3OP < inst, "CNDGT", [(set f32:$dst, (selectcc f32:$src0, FP_ZERO, f32:$src1, f32:$src2, COND_GT))] ->; +> { + let Itinerary = VecALU; +} class CNDGE_Common inst> : R600_3OP < inst, "CNDGE", [(set f32:$dst, (selectcc f32:$src0, FP_ZERO, f32:$src1, f32:$src2, COND_GE))] ->; +> { + let Itinerary = VecALU; +} let isCodeGenOnly = 1, isPseudo = 1, Namespace = "AMDGPU" in { @@ -984,35 +988,30 @@ multiclass CUBE_Common inst> { class EXP_IEEE_Common inst> : R600_1OP_Helper < inst, "EXP_IEEE", fexp2 > { - let TransOnly = 1; let Itinerary = TransALU; } class FLT_TO_INT_Common inst> : R600_1OP_Helper < inst, "FLT_TO_INT", fp_to_sint > { - let TransOnly = 1; let Itinerary = TransALU; } class INT_TO_FLT_Common inst> : R600_1OP_Helper < inst, "INT_TO_FLT", sint_to_fp > { - let TransOnly = 1; let Itinerary = TransALU; } class FLT_TO_UINT_Common inst> : R600_1OP_Helper < inst, "FLT_TO_UINT", fp_to_uint > { - let TransOnly = 1; let Itinerary = TransALU; } class UINT_TO_FLT_Common inst> : R600_1OP_Helper < inst, "UINT_TO_FLT", uint_to_fp > { - let TransOnly = 1; let Itinerary = TransALU; } @@ -1023,7 +1022,6 @@ class LOG_CLAMPED_Common inst> : R600_1OP < class LOG_IEEE_Common inst> : R600_1OP_Helper < inst, "LOG_IEEE", flog2 > { - let TransOnly = 1; let Itinerary = TransALU; } @@ -1033,72 +1031,61 @@ class ASHR_Common inst> : R600_2OP_Helper ; class MULHI_INT_Common inst> : R600_2OP_Helper < inst, "MULHI_INT", mulhs > { - let TransOnly = 1; let Itinerary = TransALU; } class MULHI_UINT_Common inst> : R600_2OP_Helper < inst, "MULHI", mulhu > { - let TransOnly = 1; let Itinerary = TransALU; } class MULLO_INT_Common inst> : R600_2OP_Helper < inst, "MULLO_INT", mul > { - let TransOnly = 1; let Itinerary = TransALU; } class MULLO_UINT_Common inst> : R600_2OP { - let TransOnly = 1; let Itinerary = TransALU; } class RECIP_CLAMPED_Common inst> : R600_1OP < inst, "RECIP_CLAMPED", [] > { - let TransOnly = 1; let Itinerary = TransALU; } class RECIP_IEEE_Common inst> : R600_1OP < inst, "RECIP_IEEE", [(set f32:$dst, (fdiv FP_ONE, f32:$src0))] > { - let TransOnly = 1; let Itinerary = TransALU; } class RECIP_UINT_Common inst> : R600_1OP_Helper < inst, "RECIP_UINT", AMDGPUurecip > { - let TransOnly = 1; let Itinerary = TransALU; } class RECIPSQRT_CLAMPED_Common inst> : R600_1OP_Helper < inst, "RECIPSQRT_CLAMPED", int_AMDGPU_rsq > { - let TransOnly = 1; let Itinerary = TransALU; } class RECIPSQRT_IEEE_Common inst> : R600_1OP < inst, "RECIPSQRT_IEEE", [] > { - let TransOnly = 1; let Itinerary = TransALU; } class SIN_Common inst> : R600_1OP < inst, "SIN", [(set f32:$dst, (SIN_HW f32:$src0))]>{ let Trig = 1; - let TransOnly = 1; let Itinerary = TransALU; } class COS_Common inst> : R600_1OP < inst, "COS", [(set f32:$dst, (COS_HW f32:$src0))]> { let Trig = 1; - let TransOnly = 1; let Itinerary = TransALU; } @@ -1480,7 +1467,6 @@ let hasSideEffects = 1 in { def FLT_TO_INT_eg : FLT_TO_INT_Common<0x50> { let Pattern = []; - let TransOnly = 0; let Itinerary = AnyALU; }