feat: Parse and verify RISCV stack operations

[nchappe]

This departs a bit from xdslproject/xdsl#6086 as alloca expects a value_type attribute to determine the size of the allocation, and it returns a RISCV register.

[github-actions]

VeIR Benchmarks

Details

Benchmark suite	Current: 85367f4	Previous: 74ee571	Ratio
add-fold-worklist/create	2226000 ns (± 106330)	1853000 ns (± 23158)	1.20
add-fold-worklist/rewrite	3768000 ns (± 297353)	3362000 ns (± 53630)	1.12
add-fold-worklist-local/create	2290000 ns (± 103468)	1865000 ns (± 91759)	1.23
add-fold-worklist-local/rewrite	3044000 ns (± 38473)	2833000 ns (± 66759)	1.07
add-zero-worklist/create	2208000 ns (± 50715)	1844000 ns (± 85301)	1.20
add-zero-worklist/rewrite	2497000 ns (± 86002)	2174000 ns (± 31092)	1.15
add-zero-reuse-worklist/create	1824500 ns (± 88529)	1558000 ns (± 10320)	1.17
add-zero-reuse-worklist/rewrite	1984500 ns (± 40153)	1767000 ns (± 20550)	1.12
mul-two-worklist/create	2216000 ns (± 33731)	1855000 ns (± 83071)	1.19
mul-two-worklist/rewrite	5228000 ns (± 64213)	4729000 ns (± 61733)	1.11
add-fold-forwards/create	2191000 ns (± 76917)	1874000 ns (± 87837)	1.17
add-fold-forwards/rewrite	3018000 ns (± 27392)	2683000 ns (± 39771)	1.12
add-zero-forwards/create	2222000 ns (± 71248)	1876500 ns (± 94595)	1.18
add-zero-forwards/rewrite	1974000 ns (± 25704)	1725000 ns (± 29016)	1.14
add-zero-reuse-forwards/create	1836000 ns (± 43365)	1554000 ns (± 47788)	1.18
add-zero-reuse-forwards/rewrite	1528000 ns (± 54684)	1499000 ns (± 69831)	1.02
mul-two-forwards/create	2201000 ns (± 104610)	1853000 ns (± 86130)	1.19
mul-two-forwards/rewrite	3650000 ns (± 75145)	3230000 ns (± 97941)	1.13
add-zero-reuse-first/create	1771000 ns (± 91141)	1540000 ns (± 45075)	1.15
add-zero-reuse-first/rewrite	8000 ns (± 1278)	9000 ns (± 408)	0.89
add-zero-lots-of-reuse-first/create	1897000 ns (± 94203)	1556000 ns (± 27499)	1.22
add-zero-lots-of-reuse-first/rewrite	782000 ns (± 27004)	750000 ns (± 12582)	1.04

This comment was automatically generated by workflow using github-action-benchmark.

[regehr]

I'm onboard with alloca here, but I don't understand why we would want stack-specific loads and stores. In general, a pointer can refer either to the stack, or the heap, or globals -- so it's not going to be easy to actually use these in practice, except in restricted situations. I would expect that the RISC-V memory model is a flat 64-bit address space and then normal loads and stores can go anywhere inside it.

[math-fehr]

But let's wait for @regehr to approve it once the comments are addressed

[nchappe]

Updated as discussed, and the stack slot type becomes unnecessary.

[regehr]

LGTM!