https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69564
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed|2016-02-01 00:00:00 |2026-2-2
--- Comment #48 from Richard Biener <rguenth at gcc dot gnu.org> ---
The C vs. C++ regression is gone (on a Zen2 uarch). Enabling -flto ontop of
-Ofast still pessimizes LU. MonteCarlo is too fast, we never get above the 2s
default, the 'int' cycles overflows to zero first.
Non-LTO:
Using 1.00 seconds min time per kenel.
Composite Score: 3545.04
FFT Mflops: 2780.77 (N=1024)
SOR Mflops: 2419.00 (100 x 100)
MonteCarlo: Mflops: 829.83
Sparse matmult Mflops: 3526.84 (N=1000, nz=5000)
LU Mflops: 8168.78 (M=100, N=100)
LTO:
Using 1.00 seconds min time per kenel.
Composite Score: 2876.78
FFT Mflops: 2813.10 (N=1024)
SOR Mflops: 2406.62 (100 x 100)
MonteCarlo: Mflops: 2264.80
Sparse matmult Mflops: 2667.71 (N=1000, nz=5000)
LU Mflops: 4231.66 (M=100, N=100)
And with -large 1.:
Non-LTO:
Using 1.00 seconds min time per kenel.
Composite Score: 3240.91
FFT Mflops: 204.51 (N=1048576)
SOR Mflops: 2216.31 (1000 x 1000)
MonteCarlo: Mflops: 825.34
Sparse matmult Mflops: 4097.13 (N=100000, nz=1000000)
LU Mflops: 8861.28 (M=1000, N=1000)
LTO:
Using 1.00 seconds min time per kenel.
Composite Score: 2006.50
FFT Mflops: 206.25 (N=1048576)
SOR Mflops: 2221.03 (1000 x 1000)
MonteCarlo: Mflops: 2264.83
Sparse matmult Mflops: 2829.64 (N=100000, nz=1000000)
LU Mflops: 2510.76 (M=1000, N=1000)
