llvm::nvvm

Helpers for NVVM / NVPTX intrinsic enums, printers, and FP semantics.

Enums

NameDescription
CTAGroupKind CTA group variants for TMA / TCGEN05 PTX instructions.
EvictPolicyType Eviction priorities for prefetch and applypriority intrinsics.
TMAReductionOp Reduction ops for TMA shared-to-global bulk tensor copies.
Tcgen05CollectorUsageOp Collector usage operations for TCGEN05 MMA.
Tcgen05MMACollectorBBuffer B-buffer selectors for TCGEN05 MMA collectors.
Tcgen05MMAKind Matrix-multiply-accumulate data kinds for TCGEN05 MMA.
TensormapElemType Element types encoded in a tensormap descriptor.
TensormapFillMode Out-of-bounds fill modes for tensormap descriptors.
TensormapInterleaveLayout Interleave layouts for tensormap descriptors.
TensormapSwizzleAtomicity Swizzle atomicity widths for tensormap descriptors.
TensormapSwizzleMode Swizzle modes for tensormap descriptors.

Functions

NameDescription
FAddShouldFTZ Return true if the NVVM add intrinsic flushes denormals to zero.
FDivShouldFTZ Return true if the NVVM div intrinsic flushes denormals to zero.
FMAShouldFTZ Return true if the NVVM FMA intrinsic flushes denormals to zero.
FMinFMaxIsXorSignAbs Return true if the fmin/fmax intrinsic uses the xorsign-abs variant.
FMinFMaxPropagatesNaNs Return true if the fmin/fmax intrinsic propagates NaN operands.
FMinFMaxShouldFTZ Return true if the fmin/fmax intrinsic flushes denormals to zero.
FMulShouldFTZ Return true if the NVVM mul intrinsic flushes denormals to zero.
FPToIntegerIntrinsicNaNZero Return true if the FP-to-integer intrinsic converts NaN inputs to zero.
FPToIntegerIntrinsicResultIsSigned Return true if the FP-to-integer intrinsic converts to a signed integer.
FPToIntegerIntrinsicShouldFTZ Return true if the FP-to-integer intrinsic flushes denormals to zero.
GetFAddRoundingMode Return the rounding mode of an NVVM floating-point add intrinsic.
GetFDivRoundingMode Return the rounding mode of an NVVM floating-point div intrinsic.
GetFMARoundingMode Return the rounding mode of an NVVM fused multiply-add intrinsic.
GetFMulRoundingMode Return the rounding mode of an NVVM floating-point mul intrinsic.
GetFPToIntegerRoundingMode Return the rounding mode of an NVVM FP-to-integer conversion intrinsic.
GetNVVMDenormMode Map an FTZ preference to the corresponding LLVM denormal mode.
GetRCPRoundingMode Return the rounding mode of an NVVM reciprocal intrinsic.
RCPShouldFTZ Return true if the reciprocal intrinsic flushes denormals to zero.
UnaryMathIntrinsicShouldFTZ Return true if the unary math intrinsic flushes denormals to zero.
getEvictPolicyName Return the PTX name for an L2 eviction policy.
getTMATensorReductionOpName Return the PTX name for a TMA tensor reduction operation.
printEvictPolicyType Print an eviction policy immediate to OS.
printTMAReductionOp Print a TMA reduction op immediate to OS.
printTcgen05CollectorUsageOp Print a TCGEN05 collector usage op immediate to OS.
printTcgen05MMACollectorBBuffer Print a TCGEN05 MMA collector B-buffer immediate to OS.
printTcgen05MMAKind Print a TCGEN05 MMA kind immediate to OS.
printTensormapElemType Print a tensormap element type immediate to OS.
printTensormapFillMode Print a tensormap fill mode immediate to OS.
printTensormapInterleaveLayout Print a tensormap interleave layout immediate to OS.
printTensormapSwizzleAtomicity Print a tensormap swizzle atomicity immediate to OS.
printTensormapSwizzleMode Print a tensormap swizzle mode immediate to OS.