This document is relevant for: Inf2, Trn1, Trn2, Trn3

NCC_EVRF062

Error message: Opcode (power) does not support 64-bit integer operands on this target; they are downcast to 32-bit integer, which loses precision for values larger than 2^31.

This error occurs when the power operation is applied to int64 operands. power has no native 64-bit integer datapath, so it is emulated by downcasting the operands to int32. Because int32 can only represent signed values up to 2^31 - 1, this downcast loses precision for larger magnitudes, so the compiler rejects the operation by default rather than silently losing precision.

Why this happens

Unlike the operations that fall back to floating-point compute (see NCC_EVRF035, NCC_EVRF060, NCC_EVRF061, and NCC_EVRF063), power runs in the integer domain by narrowing its int64 operands to int32. int32 can only represent integers exactly up to 2^31, so any result above that limit would lose precision.

Common causes

• A model raises an int64 tensor to an integer power.

Resolution

1. Explicitly cast the operands to a supported type so the operation does not require an int64 datapath.

2. Allow the int64 to int32 downcast explicitly by passing --implicit-integer-downcast=pow (or --implicit-integer-downcast=all). This downgrades the error to a warning and lets the operation run through the int64 to int32 downcast.

   Note

   Turning this on might cause numerical inaccuracy for values larger than 2^31.

This document is relevant for: Inf2, Trn1, Trn2, Trn3