TEXTRACT

Tile Operation Diagram

TEXTRACT tile operation

Introduction

Extract a sub-tile from a source tile.

Math Interpretation

Conceptually copies a window starting at (indexRow, indexCol) from src into dst. Exact mapping depends on layouts.

Let R = dst.GetValidRow() and C = dst.GetValidCol(). For 0 <= i < R and 0 <= j < C:

\[ \mathrm{dst}_{i,j} = \mathrm{src}_{\mathrm{indexRow}+i,\; \mathrm{indexCol}+j} \]

Assembly Syntax

PTO-AS form: see PTO-AS Specification.

Synchronous form:

%dst = textract %src[%r0, %r1] : !pto.tile<...> -> !pto.tile<...>

AS Level 1 (SSA)

%dst = pto.textract %src, %idxrow, %idxcol : (!pto.tile<...>, dtype, dtype) -> !pto.tile<...>

AS Level 2 (DPS)

pto.textract ins(%src, %idxrow, %idxcol : !pto.tile_buf<...>, dtype, dtype) outs(%dst : !pto.tile_buf<...>)

C++ Intrinsic

Declared in include/pto/common/pto_instr.hpp:

template <typename DstTileData, typename SrcTileData, typename... WaitEvents>
PTO_INST RecordEvent TEXTRACT(DstTileData &dst, SrcTileData &src, uint16_t indexRow = 0, uint16_t indexCol = 0, WaitEvents &... events);

template <typename DstTileData, typename SrcTileData, ReluPreMode reluMode, typename... WaitEvents>
PTO_INST RecordEvent TEXTRACT(DstTileData &dst, SrcTileData &src, uint16_t indexRow, uint16_t indexCol, WaitEvents &... events);

template <typename DstTileData, typename SrcTileData, ReluPreMode reluMode = ReluPreMode::NoRelu,
          typename... WaitEvents>
PTO_INST RecordEvent TEXTRACT(DstTileData &dst, SrcTileData &src, uint64_t preQuantScalar, uint16_t indexRow, uint16_t indexCol, WaitEvents &... events);

template <typename DstTileData, typename SrcTileData, typename FpTileData, ReluPreMode reluMode = ReluPreMode::NoRelu,
          typename... WaitEvents>
PTO_INST RecordEvent TEXTRACT_FP(DstTileData &dst, SrcTileData &src, FpTileData &fp, uint16_t indexRow, uint16_t indexCol, WaitEvents &... events);

Constraints

  • Implementation checks (A2A3):
    • DstTileData::DType must equal SrcTileData::DType and must be one of: int8_t, half, bfloat16_t, float.
    • Source fractal must satisfy: (SFractal == ColMajor && isRowMajor) or (SFractal == RowMajor && !isRowMajor). In GEMV scenarios, the source fractal satisfies (SrcTileData::Rows == 1 && SrcTileData::isRowMajor) for Left.
    • Runtime bounds checks:
    • indexRow + DstTileData::Rows <= SrcTileData::Rows
    • indexCol + DstTileData::Cols <= SrcTileData::Cols
    • Destination must be TileType::Left or TileType::Right with a target-supported fractal configuration.
  • Implementation checks (A5):
    • DstTileData::DType must equal SrcTileData::DType and must be one of: int8_t, hifloat8_t, float8_e5m2_t, float8_e4m3_t, half, bfloat16_t, float, float4_e2m1x2_t, float4_e1m2x2_t, float8_e8m0_t.
    • Source fractal must satisfy: (SFractal == ColMajor && isRowMajor) or (SFractal == RowMajor && !isRowMajor) for Left/Right, In GEMV scenarios, the source fractal satisfies (SrcTileData::Rows == 1 && SrcTileData::isRowMajor) for Left. (SFractal == RowMajor && isRowMajor) for ScaleLeft, (SFractal == ColMajor && !isRowMajor) for ScaleRight.
    • Destination supports Mat -> Left/Right/Scale, Acc -> Mat (including relu/scalar-quant/vector-quant forms), and also supports specific Vec -> Mat extraction paths.
    • The vector-quantized form additionally requires an FpTileData scaling operand, matching the TEXTRACT_FP(...) interface shown above.

Examples

Auto

#include <pto/pto-inst.hpp>

using namespace pto;

void example_auto() {
  using SrcT = Tile<TileType::Mat, float, 16, 16, BLayout::RowMajor, 16, 16, SLayout::ColMajor>;
  using DstT = TileLeft<float, 16, 16>;
  SrcT src;
  DstT dst;
  TEXTRACT(dst, src, /*indexRow=*/0, /*indexCol=*/0);
}

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 src;
  DstT dst;
  TASSIGN(src, 0x1000);
  TASSIGN(dst, 0x2000);
  TEXTRACT(dst, src, /*indexRow=*/0, /*indexCol=*/0);
}

ASM Form Examples

Auto Mode

# Auto mode: compiler/runtime-managed placement and scheduling.
%dst = pto.textract %src, %idxrow, %idxcol : (!pto.tile<...>, dtype, dtype) -> !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.textract %src, %idxrow, %idxcol : (!pto.tile<...>, dtype, dtype) -> !pto.tile<...>

PTO Assembly Form

%dst = textract %src[%r0, %r1] : !pto.tile<...> -> !pto.tile<...>
# AS Level 2 (DPS)
pto.textract ins(%src, %idxrow, %idxcol : !pto.tile_buf<...>, dtype, dtype) outs(%dst : !pto.tile_buf<...>)