The GitHub Actions job "Lint" on tvm.git/fix/cuda-min-max-nan has failed. Run started by GitHub user LngelKyo (triggered by LngelKyo).
Head commit for run: cbe6a062df6c3c37b0499853553c386c1606e0f7 / LngelKyo <[email protected]> [Fix][CUDA] Constant-operand fast paths for NaN-preserving min/max Match #20054's four constant-operand cases (host side, min_max_utils.h): - lhs is a constant NaN (FloatImm or Broadcast of FloatImm): printed directly as the surviving operand; - lhs is a constant non-NaN: ordered compare, b's NaN makes the compare false and b is kept; - rhs is a constant NaN: ((a != a) ? a : b) keeps a on any outcome; - rhs is a constant non-NaN: reversed non-strict compare (< -> >=, > -> <=) so a's NaN makes the compare false and a survives; ties take b. The constant side is not SSA-bound (no side effects); the non-constant side is bound once inside its own scope, matching the chained-statement fix. Helper AsFloatImm (FloatImm or Broadcast of FloatImm) lives in the anonymous namespace of codegen_cuda.cc, noted as sharing semantics with #20054's min_max_utils.h so the two can be merged later. Tests mirror #20054's test_min_max_float_imm_operand: data [nan, -0.0, 0.0, -1.0, 1.0, inf, -inf, 2.0], parameterized op (2) x const_side (2) x const in {0.0, NaN} (2) x form {scalar, vec4} (2) = 32 cases, bitwise against np.where(cmp | isnan(lhs), lhs, rhs). Report URL: https://github.com/apache/tvm/actions/runs/35669401104 With regards, GitHub Actions via GitBox --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
