Adapt to GPUCompiler 2.13's atomics handling

[maleadt]

Follow-up to #1143, for GPUCompiler 2.13:

• GPUCompiler 2.13 validates GCN atomics and rejects synchronization scopes the back-end doesn't know, so test/core/codegen.jl on main now errors on the test that checks an unknown scope gets no memory assumptions. That case is now compiled without IR validation.
• Let atomics assume coarse-grained, local memory unless they use the system scope #1143 left the memory assumptions off flat usub_sat, because LLVM 22 couldn't select the native instruction on gfx10.3 and gfx11 with them ([AMDGPU] Cannot select AtomicLoadUSubSat on LDS (gfx10.3/gfx11) and flat memory (gfx12, or with no.fine.grained.memory) llvm/llvm-project#229442). GPUCompiler 2.13 handles that: it requires AMDGPU_LLVM_Backend_jll 23.1.3, which has the fix, and expands the operation to a compare-and-swap loop for Julia's own LLVM 22+ (Validate atomics for the GCN target, and work around two LLVM atomic bugs GPUCompiler.jl#997). So this requires GPUCompiler 2.13 and treats usub_sat like the other operations.

Tested core/codegen on Julia 1.13 (gfx1036 host, offline gfx1030/gfx90a code generation) with GPUCompiler 2.13.0 and AMDGPU_LLVM_Backend_jll 23.1.3: main fails with one error, both commits pass.

[maleadt]

Merging this a little quicker since this fixes main again.

[github-actions]

AMDGPU.jl Benchmarks

Details

Benchmark suite	Current: d4e09ad	Previous: 6f37fdf	Ratio
amdgpu/synchronization/context/device	547.5 ns	547.5 ns	1
amdgpu/synchronization/stream/blocking	222.5 ns	222.5 ns	1
amdgpu/synchronization/stream/nonblocking	312.5 ns	305 ns	1.02
applications/bitonic_sort	1111449.5 ns	1121018 ns	0.99
applications/convolution	90956.25 ns	91856.25 ns	0.99
applications/floyd_warshall	9092236.75 ns	9088943 ns	1.00
applications/histogram	793914.75 ns	795198.5 ns	1.00
applications/prefix_sum	188477.75 ns	188172.75 ns	1.00
array/accumulate/Float32/1d	82211.25 ns	82703.5 ns	0.99
array/accumulate/Float32/dims=1	266236.75 ns	259653.5 ns	1.03
array/accumulate/Float32/dims=1L	100556.5 ns	91916.25 ns	1.09
array/accumulate/Float32/dims=2	75706 ns	105999 ns	0.71
array/accumulate/Float32/dims=2L	3030870 ns	3029801.75 ns	1.00
array/accumulate/Int64/1d	86631.25 ns	93226.5 ns	0.93
array/accumulate/Int64/dims=1	241926.25 ns	242948.25 ns	1.00
array/accumulate/Int64/dims=1L	111059.25 ns	112674.25 ns	0.99
array/accumulate/Int64/dims=2	100699 ns	80158.75 ns	1.26
array/accumulate/Int64/dims=2L	3390968.25 ns	3392649.25 ns	1.00
array/broadcast	28790.25 ns	29063 ns	0.99
array/construct	2365 ns	2220 ns	1.07
array/copy	38015.75 ns	38870.5 ns	0.98
array/copyto!/cpu_to_gpu	89096.25 ns	88961.25 ns	1.00
array/copyto!/gpu_to_cpu	89319 ns	89613.75 ns	1.00
array/copyto!/gpu_to_gpu	33700.5 ns	34700.5 ns	0.97
array/iteration/findall/bool	121814.5 ns	119901.75 ns	1.02
array/iteration/findall/int	112469 ns	112736.5 ns	1.00
array/iteration/findfirst/bool	151912.5 ns	152529.75 ns	1.00
array/iteration/findfirst/int	153660 ns	153119.5 ns	1.00
array/iteration/findmin/1d	103469 ns	104196.25 ns	0.99
array/iteration/findmin/2d	93824 ns	93761.25 ns	1.00
array/iteration/logical	182895.25 ns	183702.5 ns	1.00
array/iteration/scalar	299422.25 ns	309796.75 ns	0.97
array/permutedims/2d	61138.25 ns	59476 ns	1.03
array/permutedims/3d	56328.5 ns	57221 ns	0.98
array/permutedims/4d	41290.5 ns	43595.5 ns	0.95
array/random/rand/Float32	44733.25 ns	45258 ns	0.99
array/random/rand/Int64	42983.25 ns	37798 ns	1.14
array/random/rand!/Float32	41388.25 ns	41948 ns	0.99
array/random/rand!/Int64	31733 ns	37215.5 ns	0.85
array/random/randn/Float32	68153.5 ns	72228.25 ns	0.94
array/random/randn!/Float32	53348.25 ns	54075.75 ns	0.99
array/reductions/mapreduce/Float32/1d	107961.5 ns	108671.5 ns	0.99
array/reductions/mapreduce/Float32/dims=1	87391.25 ns	88146.25 ns	0.99
array/reductions/mapreduce/Float32/dims=1L	835465.5 ns	836306.5 ns	1.00
array/reductions/mapreduce/Float32/dims=2	92706.5 ns	94228.75 ns	0.98
array/reductions/mapreduce/Float32/dims=2L	135524.5 ns	135486.75 ns	1.00
array/reductions/mapreduce/Int64/1d	106111.75 ns	108869 ns	0.97
array/reductions/mapreduce/Int64/dims=1	88076.5 ns	88541.25 ns	0.99
array/reductions/mapreduce/Int64/dims=1L	838898 ns	841141.75 ns	1.00
array/reductions/mapreduce/Int64/dims=2	91298.75 ns	92663.75 ns	0.99
array/reductions/mapreduce/Int64/dims=2L	136294.75 ns	136469.5 ns	1.00
array/reductions/reduce/Float32/1d	100604 ns	108931.5 ns	0.92
array/reductions/reduce/Float32/dims=1	87638.75 ns	88088.5 ns	0.99
array/reductions/reduce/Float32/dims=1L	834180.5 ns	835606.5 ns	1.00
array/reductions/reduce/Float32/dims=2	93304 ns	94116.25 ns	0.99
array/reductions/reduce/Float32/dims=2L	135417 ns	135959.5 ns	1.00
array/reductions/reduce/Int64/1d	108276.75 ns	107759 ns	1.00
array/reductions/reduce/Int64/dims=1	87876.25 ns	87166.25 ns	1.01
array/reductions/reduce/Int64/dims=1L	844910.5 ns	837371.5 ns	1.01
array/reductions/reduce/Int64/dims=2	92096.25 ns	92491.25 ns	1.00
array/reductions/reduce/Int64/dims=2L	135722 ns	135859.25 ns	1.00
array/reverse/1d	42750.75 ns	35203 ns	1.21
array/reverse/1dL	68171 ns	69038.25 ns	0.99
array/reverse/1dL_inplace	53395.75 ns	54128.25 ns	0.99
array/reverse/1d_inplace	37080.5 ns	37770.5 ns	0.98
array/reverse/2d	44695.75 ns	46583 ns	0.96
array/reverse/2dL	89976.25 ns	91263.75 ns	0.99
array/reverse/2dL_inplace	48435.75 ns	65028.5 ns	0.74
array/reverse/2d_inplace	35360.5 ns	37158 ns	0.95
array/sorting/1d	319555 ns	330742 ns	0.97
gemm/tiled	1941872.25 ns	1949946.75 ns	1.00
gemm/tiled_unbounded	1984465.5 ns	2914637.75 ns	0.68
integration/byval/reference	39611 ns	39901 ns	0.99
integration/byval/slices=1	40370 ns	39770 ns	1.02
integration/byval/slices=2	158342 ns	157742 ns	1.00
integration/byval/slices=3	238014 ns	223213 ns	1.07
integration/volumerhs	4877064 ns	4883697 ns	1.00
kernel/indexing	29825.5 ns	28847.75 ns	1.03
kernel/indexing_checked	36175.5 ns	32488 ns	1.11
kernel/launch	1212.5 ns	1220 ns	0.99
kernel/rand	50373.25 ns	51218.25 ns	0.98
latency/import	1633802228 ns	1628061194 ns	1.00
latency/precompile	29177559763 ns	29168346331 ns	1.00
latency/ttfp	2186237051 ns	2181953516 ns	1.00
stencil/diffusion3d	1585699.25 ns	1583662 ns	1.00
stencil/diffusion3d_checked	1631142.5 ns	1637802.75 ns	1.00

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