[llvm] [MCA] Extend -instruction-tables option with verbosity levels (PR #130574)

[Min-Yih Hsu via llvm-commits]

================
@@ -141,6 +241,33 @@ void InstructionInfoView::collectData(
     IIVDEntry.mayLoad = MCDesc.mayLoad();
     IIVDEntry.mayStore = MCDesc.mayStore();
     IIVDEntry.hasUnmodeledSideEffects = MCDesc.hasUnmodeledSideEffects();
+
+    if (PrintFullInfo) {
+      // Get latency with bypass
+      IIVDEntry.Bypass =
+          IIVDEntry.Latency - MCSchedModel::getBypassDelayCycles(STI, SCDesc);
+      IIVDEntry.OpcodeName = MCII.getName(Inst.getOpcode());
+      raw_string_ostream TempStream(IIVDEntry.Resources);
+      const MCWriteProcResEntry *Index = STI.getWriteProcResBegin(&SCDesc);
+      const MCWriteProcResEntry *Last = STI.getWriteProcResEnd(&SCDesc);
+      auto Sep = "";
+      for (; Index != Last; ++Index) {
+        if (!Index->ReleaseAtCycle)
+          continue;
+        const MCProcResourceDesc *MCProc =
+            SM.getProcResource(Index->ProcResourceIdx);
+        if (Index->ReleaseAtCycle > 1) {
----------------
mshockwave wrote:

let's use `vlm.v` you pointed out as an example, the timeline of two `vlm.v` -- X and Y -- looks like this:
```
          |  0  |  1  |  2  |  3  |  4  |
SiFive7CQ    X           Y
SiFive7VL          X     X     Y     Y
```
(AcquireAtCycle & ReleaseACycle has a relationship of `[AcquireAtCycle, ReleaseACycle)`)
It takes 5 cycles to execute 2 instructions, so the rthroughput would be 2.5 cycles as opposed to 3 cycles if we only consider ReleaseAtCycle.
Furthermore, if we try to execute three of these instructions, it takes a total of 7 cycles, which means the rthroughput would be 7 / 3 = 2.3333

https://github.com/llvm/llvm-project/pull/130574