[AMDGPU] Fix off by one in assert

[Flakebi]

Summary

D89217 did not subtract one when accessing SubRegFromChannelTable in one
place.

[arsenm]

Testcase?

[critson]

LGTM.

Thanks, I was going to make the same change, but you beat me to it.
Slight nit. since we use Width - 1 three times, for readability I think we should just declare a new variable for it (TableIdx?).

We cannot easily have a test case for this as it is based on TableGen data.
I would have expected this assertion to be failing already due to the error, but looking at the code generation .size() for this type is a compile time constant, so in practice this becomes assert(Offset < 32), with the index completely disregarded.

[Flakebi]

I don’t see a way to add a test case. It fails an assertion on Windows when compiling with msvc.

Changed to use TableIdx.

[critson]

In D89804#2343814, @Flakebi wrote:

Changed to use TableIdx.

Looks good, thanks!

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

for (auto &Row : SubRegFromChannelTable)
       Row.fill(AMDGPU::NoSubRegister);
     for (uint16_t Idx = 1; Idx < getNumSubRegIndices(); ++Idx) {
       unsigned Width = AMDGPUSubRegIdxRanges[Idx].Size / 32;
       unsigned Offset = AMDGPUSubRegIdxRanges[Idx].Offset / 32;
       assert(Width < SubRegFromChannelTableWidthMap.size());
       Width = SubRegFromChannelTableWidthMap[Width];
       if (Width == 0)
         continue;
-      assert((Width - 1) < SubRegFromChannelTable.size());
-      assert(Offset < SubRegFromChannelTable[Width].size());
-      SubRegFromChannelTable[Width - 1][Offset] = Idx;
+      unsigned TableIdx = Width - 1;
+      assert(TableIdx < SubRegFromChannelTable.size());
+      assert(Offset < SubRegFromChannelTable[TableIdx].size());
+      SubRegFromChannelTable[TableIdx][Offset] = Idx;
     }
   };
 
   llvm::call_once(InitializeRegSplitPartsFlag, InitializeRegSplitPartsOnce);
   llvm::call_once(InitializeSubRegFromChannelTableFlag,
                   InitializeSubRegFromChannelTableOnce);
 }