TMOV¶
Tile Operation Diagram¶
Introduction¶
Move/copy between tiles, optionally applying implementation-defined conversion modes selected by template parameters and overloads.
TMOV is used for:
- Vec -> Vec moves
- Mat -> Left/Right/Bias/Scaling/Scale(Microscaling) moves (target-dependent)
- Acc -> Vec moves (target-dependent)
Math Interpretation¶
Conceptually copies or transforms elements from src into dst over the valid region. Exact transformation depends on the selected mode and target.
For the pure copy case:
\[ \mathrm{dst}_{i,j} = \mathrm{src}_{i,j} \]
Assembly Syntax¶
PTO-AS form: see PTO-AS Specification.
The PTO AS design recommends splitting TMOV into a family of ops:
%left = tmov.m2l %mat : !pto.tile<...> -> !pto.tile<...>
%right = tmov.m2r %mat : !pto.tile<...> -> !pto.tile<...>
%bias = tmov.m2b %mat : !pto.tile<...> -> !pto.tile<...>
%scale = tmov.m2s %mat : !pto.tile<...> -> !pto.tile<...>
%vec = tmov.a2v %acc : !pto.tile<...> -> !pto.tile<...>
%v1 = tmov.v2v %v0 : !pto.tile<...> -> !pto.tile<...>
IR Level 1 (SSA)¶
%dst = pto.tmov.s2d %src : !pto.tile<...> -> !pto.tile<...>
IR Level 2 (DPS)¶
pto.tmov ins(%src : !pto.tile_buf<...>) outs(%dst : !pto.tile_buf<...>)
C++ Intrinsic¶
Declared in include/pto/common/pto_instr.hpp and include/pto/common/constants.hpp:
template <typename DstTileData, typename SrcTileData, typename... WaitEvents>
PTO_INST RecordEvent TMOV(DstTileData& dst, SrcTileData& src, WaitEvents&... events);
template <typename DstTileData, typename SrcTileData, ReluPreMode reluMode, typename... WaitEvents>
PTO_INST RecordEvent TMOV(DstTileData& dst, SrcTileData& src, WaitEvents&... events);
template <typename DstTileData, typename SrcTileData, AccToVecMode mode, ReluPreMode reluMode = ReluPreMode::NoRelu,
typename... WaitEvents>
PTO_INST RecordEvent TMOV(DstTileData& dst, SrcTileData& src, WaitEvents&... events);
template <typename DstTileData, typename SrcTileData, ReluPreMode reluMode = ReluPreMode::NoRelu,
typename... WaitEvents>
PTO_INST RecordEvent TMOV(DstTileData& dst, SrcTileData& src, uint64_t preQuantScalar, WaitEvents&... events);
template <typename DstTileData, typename SrcTileData, AccToVecMode mode, ReluPreMode reluMode = ReluPreMode::NoRelu,
typename... WaitEvents>
PTO_INST RecordEvent TMOV(DstTileData& dst, SrcTileData& src, uint64_t preQuantScalar, WaitEvents&... events);
template <typename DstTileData, typename SrcTileData, typename FpTileData, ReluPreMode reluMode = ReluPreMode::NoRelu,
typename... WaitEvents>
PTO_INST RecordEvent TMOV_FP(DstTileData& dst, SrcTileData& src, FpTileData& fp, WaitEvents&... events);
template <typename DstTileData, typename SrcTileData, typename FpTileData, AccToVecMode mode,
ReluPreMode reluMode = ReluPreMode::NoRelu, typename... WaitEvents>
PTO_INST RecordEvent TMOV(DstTileData& dst, SrcTileData& src, FpTileData& fp, WaitEvents&... events);
Constraints¶
- Implementation checks (A2A3):
- Shapes must match:
SrcTileData::Rows == DstTileData::RowsandSrcTileData::Cols == DstTileData::Cols. - Supported location pairs (compile-time checked):
Mat -> Left/Right/Bias/ScalingVec -> VecAcc -> Mat(including optional pre-quant / relu / fp variants via overloads)
- For
Acc -> Mat, additional fractal/type constraints are enforced (e.g.,Accuses NZ-like fractal,Matuses 512B fractal, and only specific dtype conversions are allowed). - Implementation checks (A5):
- For
Mat -> *, shapes must match; for someVecmoves, the effective copy size is the min of src/dst valid rows/cols. - Supported location pairs include (target-dependent):
Mat -> Left/Right/Bias/Scaling/ScaleVec -> VecandVec -> MatAcc -> VecandAcc -> Mat(including optional pre-quant / relu / fp variants via overloads)
- For
Mat -> Left/Right, additional fractal and dtype constraints are enforced viaCommonCheck(source fractal must be compatible and element types must match). - For
Acc -> Vec/Mat, additional fractal/type/alignment constraints are enforced viaCheckTMovAccValid. - For
Mat -> Scale, additional fractal and dtype constraints are enforced viaCommonCheckMX(source fractal must be compatible and element types must match).
Examples¶
Auto¶
#include <pto/pto-inst.hpp>
using namespace pto;
void example_auto() {
using TileT = Tile<TileType::Vec, float, 16, 16>;
TileT src, dst;
TMOV(dst, src);
}
Manual¶
#include <pto/pto-inst.hpp>
using namespace pto;
void example_manual() {
using SrcT = Tile<TileType::Mat, float, 16, 16, BLayout::RowMajor, 16, 16, SLayout::ColMajor>;
using DstT = TileLeft<float, 16, 16>;
SrcT mat;
DstT left;
TASSIGN(mat, 0x1000);
TASSIGN(left, 0x2000);
TMOV(left, mat);
}
ASM Form Examples¶
Auto Mode¶
# Auto mode: compiler/runtime-managed placement and scheduling.
%dst = pto.tmov.s2d %src : !pto.tile<...> -> !pto.tile<...>
Manual Mode¶
# Manual mode: bind resources explicitly before issuing the instruction.
# Optional for tile operands:
# pto.tassign %arg0, @tile(0x1000)
# pto.tassign %arg1, @tile(0x2000)
%dst = pto.tmov.s2d %src : !pto.tile<...> -> !pto.tile<...>
PTO Assembly Form¶
%dst = pto.tmov.s2d %src : !pto.tile<...> -> !pto.tile<...>
# IR Level 2 (DPS)
pto.tmov ins(%src : !pto.tile_buf<...>) outs(%dst : !pto.tile_buf<...>)