TRELU

Instruction Diagram

Introduction

Element-wise ReLU of a tile.

Mathematical Semantics

For each element (i, j) in the valid region:

$$\mathrm{dst}_{i,j} = \max(\mathrm{src}_{i,j}, 0)$$

Assembly Syntax

Synchronous form:

%dst = trelu %src : !pto.tile<...>

AS Level 1 (SSA)

%dst = pto.trelu %src : !pto.tile<...> -> !pto.tile<...>

AS Level 2 (DPS)

pto.trelu ins(%src : !pto.tile_buf<...>) outs(%dst : !pto.tile_buf<...>)

C++ Built-in APIs

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

The public include header is <pto/pto-inst.hpp>, and the internal declaration is located in pto/common/pto_instr.hpp.

template <typename TileDataDst, typename TileDataSrc, typename... WaitEvents>
PTO_INST RecordEvent TRELU(TileDataDst &dst, TileDataSrc &src, WaitEvents &... events);

Constraints

• Implementation check (Atlas A2/A3 training products/Atlas A2/A3 inference products):
  • TileData::DType must be one of the following: half, float, int32_t.
  • The tile layout must be row-major (TileData::isRowMajor).
  • The tile position must be a vector (TileData::Loc == TileType::Vec).
  • Static valid boundary: TileData::ValidRow <= TileData::Rows and TileData::ValidCol <= TileData::Cols.
  • Runtime: the src and dst tiles must have the same validRow/validCol.
• Implementation check (Ascend 950PR/Ascend 950DT):
  • TileData::DType must be one of the following: half, float, int32_t (consistent with Atlas A2/A3 training products/Atlas A2/A3 inference products).
  • The tile layout must be row-major (TileData::isRowMajor).
  • The tile position must be a vector (TileData::Loc == TileType::Vec).
  • Static valid boundary: TileData::ValidRow <= TileData::Rows and TileData::ValidCol <= TileData::Cols.
  • At runtime: the src and dst tiles should have the same validRow/validCol.
• Valid region:
  • This operation uses dst.GetValidRow()/dst.GetValidCol() as the iteration domain; src/dst are assumed to be compatible (not verified through an explicit runtime check in this operation).

Examples

#include <pto/pto-inst.hpp>

using namespace pto;

void example() {
  using TileT = Tile<TileType::Vec, float, 16, 16>;
  TileT x, out;
  TRELU(out, x);
}

ASM Examples

Automatic Mode

# Automatic mode: the compiler/runtime handles resource placement and scheduling.
%dst = pto.trelu %src : !pto.tile<...> -> !pto.tile<...>

Manual Mode

# Manual mode: explicitly bind resources first, then issue the instruction.
# Optional (when the instruction contains tile operands):
# pto.tassign %arg0, @tile(0x1000)
# pto.tassign %arg1, @tile(0x2000)
%dst = pto.trelu %src : !pto.tile<...> -> !pto.tile<...>

PTO Assembly Form

%dst = trelu %src : !pto.tile<...>
# AS Level 2 (DPS)
pto.trelu ins(%src : !pto.tile_buf<...>) outs(%dst : !pto.tile_buf<...>)