[llvm] TargetSchedule: correct latency by cycles elapsed from def to use (PR #74088)

[Ramkumar Ramachandra via llvm-commits]

artagnon wrote:

The benchmark results on rv64 without the vector extension are as follows:

```
Program                                       upstream patch   diff
test-suite...External/Embench/nsichneu.test   2544804  2544804  0.0%
test-suite...ternal/Embench/aha-mont64.test   1672896  1672896  0.0%
test-suite...External/Embench/wikisort.test    503171   503171  0.0%
test-suite :: External/Embench/ud.test         616379   616379  0.0%
test-suite... External/Embench/tarfind.test    942843   942843  0.0%
test-suite...xternal/Embench/statemate.test    137571   137571  0.0%
test-suite :: External/Embench/st.test         195360   195360  0.0%
test-suite... :: External/Embench/slre.test   2024077  2024077  0.0%
test-suite...al/Embench/sglib-combined.test   2601254  2601254  0.0%
test-suite...ternal/Embench/primecount.test   3582414  3582414  0.0%
test-suite...External/Embench/picojpeg.test   2340577  2340577  0.0%
test-suite...marks/Dhrystone/dhrystone.test    103635   103635  0.0%
test-suite...ternal/Embench/nettle-aes.test   2385040  2385040  0.0%
test-suite...:: External/Embench/nbody.test     81295    81295  0.0%
test-suite...: External/Embench/minver.test    614317   614317  0.0%
test-suite...: External/Embench/md5sum.test   1239777  1239777  0.0%
test-suite...xternal/Embench/huffbench.test   1949283  1949283  0.0%
test-suite :: External/Embench/edn.test       2958036  2958036  0.0%
test-suite...:: External/Embench/cubic.test    153440   153440  0.0%
test-suite...:: External/Embench/crc32.test   2962451  2962451  0.0%
test-suite...ernal/Embench/matmult-int.test   1128607  1127676 -0.1%
test-suite...xternal/CoreMark/coremark.test   1277202  1275331 -0.1%
test-suite...nal/Embench/nettle-sha256.test   1786605  1780935 -0.3%
test-suite... External/Embench/qrduino.test   2400183  2391428 -0.4%
Geomean difference                                             -0.0%
```

Questions:
1. Is the idea behind the patch sound?
2. Are all the test changes indeed improvements, with minimal/no regressions?
3. If the idea behind the patch is sound, is there a more principled way of picking the `MulFactor`?
4. If the idea behind the patch is unsound, did I just get lucky with my benchmark set on rv64?

https://github.com/llvm/llvm-project/pull/74088