Switch to CompilerCaching.jl.

[maleadt]

No description provided.

[github-actions]

AMDGPU.jl Benchmarks

Details

Benchmark suite	Current: 825b4f4	Previous: e73cebd	Ratio
amdgpu/synchronization/context/device	600 ns	600 ns	1
amdgpu/synchronization/stream/blocking	250 ns	250 ns	1
amdgpu/synchronization/stream/nonblocking	340 ns	340 ns	1
array/accumulate/Float32/1d	84321 ns	86591 ns	0.97
array/accumulate/Float32/dims=1	263584 ns	344845 ns	0.76
array/accumulate/Float32/dims=1L	135992 ns	136302 ns	1.00
array/accumulate/Float32/dims=2	134562 ns	138422 ns	0.97
array/accumulate/Float32/dims=2L	2808191 ns	2811150 ns	1.00
array/accumulate/Int64/1d	100322 ns	97821 ns	1.03
array/accumulate/Int64/dims=1	251214 ns	255884 ns	0.98
array/accumulate/Int64/dims=1L	167193 ns	163062 ns	1.03
array/accumulate/Int64/dims=2	127522 ns	126152 ns	1.01
array/accumulate/Int64/dims=2L	2990593 ns	2988132 ns	1.00
array/broadcast	147542 ns	91501 ns	1.61
array/construct	1720 ns	1740 ns	0.99
array/copy	37080 ns	39290 ns	0.94
array/copyto!/cpu_to_gpu	183423 ns	183122 ns	1.00
array/copyto!/gpu_to_cpu	183593 ns	115292 ns	1.59
array/copyto!/gpu_to_gpu	130912 ns	130942 ns	1.00
array/iteration/findall/bool	179423 ns	181202 ns	0.99
array/iteration/findall/int	201183 ns	199323 ns	1.01
array/iteration/findfirst/bool	118762 ns	125122 ns	0.95
array/iteration/findfirst/int	115992 ns	118152 ns	0.98
array/iteration/findmin/1d	171413 ns	170723 ns	1.00
array/iteration/findmin/2d	156852 ns	156842 ns	1.00
array/iteration/logical	353785 ns	348555 ns	1.02
array/iteration/scalar	290154 ns	289374 ns	1.00
array/permutedims/2d	75681 ns	74621 ns	1.01
array/permutedims/3d	75781 ns	74771 ns	1.01
array/permutedims/4d	77541 ns	77081 ns	1.01
array/random/rand/Float32	52271 ns	52431 ns	1.00
array/random/rand/Int64	59721 ns	58791 ns	1.02
array/random/rand!/Float32	91072 ns	126382 ns	0.72
array/random/rand!/Int64	144952 ns	90851 ns	1.60
array/random/randn/Float32	89341 ns	88321 ns	1.01
array/random/randn!/Float32	145872 ns	114642 ns	1.27
array/reductions/mapreduce/Float32/1d	134652 ns	133232 ns	1.01
array/reductions/mapreduce/Float32/dims=1	97062 ns	95322 ns	1.02
array/reductions/mapreduce/Float32/dims=1L	777951 ns	778031 ns	1.00
array/reductions/mapreduce/Float32/dims=2	99652 ns	97481 ns	1.02
array/reductions/mapreduce/Float32/dims=2L	298954 ns	303054 ns	0.99
array/reductions/mapreduce/Int64/1d	138872 ns	133642 ns	1.04
array/reductions/mapreduce/Int64/dims=1	97092 ns	96002 ns	1.01
array/reductions/mapreduce/Int64/dims=1L	783081 ns	783691 ns	1.00
array/reductions/mapreduce/Int64/dims=2	98412 ns	94871 ns	1.04
array/reductions/mapreduce/Int64/dims=2L	303295 ns	295344 ns	1.03
array/reductions/reduce/Float32/1d	136602 ns	133712 ns	1.02
array/reductions/reduce/Float32/dims=1	97391 ns	94911 ns	1.03
array/reductions/reduce/Float32/dims=1L	776961 ns	775850 ns	1.00
array/reductions/reduce/Float32/dims=2	99412 ns	96732 ns	1.03
array/reductions/reduce/Float32/dims=2L	301274 ns	296144 ns	1.02
array/reductions/reduce/Int64/1d	134532 ns	133632 ns	1.01
array/reductions/reduce/Int64/dims=1	96821 ns	95521 ns	1.01
array/reductions/reduce/Int64/dims=1L	784081 ns	785631 ns	1.00
array/reductions/reduce/Int64/dims=2	97931 ns	96781 ns	1.01
array/reductions/reduce/Int64/dims=2L	297474 ns	300024 ns	0.99
array/reverse/1d	44471 ns	43090 ns	1.03
array/reverse/1dL	76841 ns	74821 ns	1.03
array/reverse/1dL_inplace	134241 ns	118171 ns	1.14
array/reverse/1d_inplace	140762 ns	73531 ns	1.91
array/reverse/2d	53031 ns	51091 ns	1.04
array/reverse/2dL	102711 ns	101392 ns	1.01
array/reverse/2dL_inplace	118582 ns	178633 ns	0.66
array/reverse/2d_inplace	60341 ns	143843 ns	0.42
array/sorting/1d	345845 ns	341594 ns	1.01
integration/byval/reference	39590 ns	39160 ns	1.01
integration/byval/slices=1	40321 ns	41701 ns	0.97
integration/byval/slices=2	160092 ns	157872 ns	1.01
integration/byval/slices=3	239264 ns	240533 ns	0.99
integration/volumerhs	5043483 ns	4941880 ns	1.02
kernel/indexing	65581 ns	121521 ns	0.54
kernel/indexing_checked	132762 ns	63651 ns	2.09
kernel/launch	1570 ns	1340 ns	1.17
kernel/rand	197312 ns	195973 ns	1.01
latency/import	1740434752 ns	1591443328 ns	1.09
latency/precompile	37300401844 ns	36454927171 ns	1.02
latency/ttfp	5733290610 ns	2161911248 ns	2.65

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

[maleadt]

Done on my end. I'll leave it to AMDGPU.jl maintainers to merge this.

[luraess]

@wsmoses we have this now failing for Enzyme tests as this PR requires now GPUCompiler v2 and Enzyme is still at 1.6.2 (see https://buildkite.com/julialang/amdgpu-dot-jl/builds/3886#019f5b1f-0fd3-4bc7-8c10-878504fd0dc9/L669). Could we bump compat and patch Enzyme?

[wsmoses]

@vchuravy has a pr starting to look into it.

That said, @maleadt as the changes here don't look huge, is it possible to conditionally gate this with something like

if gpucompilerver >= 2
a()
else
b()
end

that way downstream packages that haven't adapted yet [like enzyme/reactant] can still run ci with?

And the same for cuda/metal?

[maleadt]

I'd rather not. I've added enough cruft already in GPUCompiler just to keep supporting 1.10 for Enzyme, and I don't want that leak into the back-ends too.

[wsmoses]

Fair enough, in which case would you be able to help us adapt enzyme/reactant to gpucompiler 2?