lower pow(x,1) to x

[liqiangxl]

Partially solve #3629

Currently, genPowerWithMul() only handles case with factor of 2 & 3, for y = pow(x, 1.0), it should be lowered to y = x which is much faster than the power version.

[liqiangxl]

!test --pybench

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit 0b51ec7)

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 No relevant tests

⚡ Recommended focus areas for review Code Duplication The new code has duplicated logic for handling different values of exponent. This could be simplified and made more efficient. for (int i = 0; i < exponent; ++i) { if (i != 0) { code_ << " * "; } code_ << lhs; } } else { indent() << gen(bop->out()); if (bop->out()->isScalar()) { for (int i = 0; i < exponent; ++i) { if (i == 0) { code_ << " = " << lhs; } else { code_ << " * " << lhs; } } } else { for (int i = 0; i < exponent; ++i) { if (i == 0) { code_ << "\n"; indent() << kTab << "= " << lhs; } else { indent() << "\n" << kTab << "* " << lhs; } } } } Missing Comments The new code lacks comments explaining the purpose of the changes and the logic behind them. // Only **1, **2 and **3 are considered if (!(exponent == 1 || exponent == 2 || exponent == 3)) { return false; } auto lhs = gen(bop->lhs()); if (print_inline_) { for (int i = 0; i < exponent; ++i) { if (i != 0) { code_ << " * "; } code_ << lhs; } } else { indent() << gen(bop->out()); if (bop->out()->isScalar()) { for (int i = 0; i < exponent; ++i) { if (i == 0) { code_ << " = " << lhs; } else { code_ << " * " << lhs; } } } else { for (int i = 0; i < exponent; ++i) { if (i == 0) { code_ << "\n"; indent() << kTab << "= " << lhs; } else { indent() << "\n" << kTab << "* " << lhs; } } } }

[naoyam]

Code duplication is a bit concerning. Can't you do some simplification? Something like:

for (int i = 0; i < exponent; ++I) {
  if (i != 0) {
    code_ << " * ";
  }
   code_ << lhs;
}

[liqiangxl]

@Priya2698 do you know how to run thunder benchmark? I didn't find any thunder benchmark results in dashboard generated from this PR using !test --pybench

[naoyam]

LGTM

[liqiangxl]

!test --pybench --dev

[liqiangxl]

!build