Test @fastmath integer powers

[maleadt]

@fastmath x^n with an integer n emits llvm.powi, which Apple's back-end can't compile, so the kernel fails with "Compilation to native code failed". JuliaGPU/GPUCompiler.jl#946 expands the intrinsic into multiplies, in the same order the CPU uses.

This PR bumps GPUCompiler to 2.9 and adds an integration test. The test compares @fastmath x^n on the GPU with the CPU result, bit for bit, for special bases (±0, ±Inf, NaN) and exponents including typemin/typemax(Int32). It covers run-time and constant exponents, plus Float16. With the current GPUCompiler it fails with the compilation error above.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 86.47%. Comparing base (26c9e7d) to head (162dfdf).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #978      +/-   ##
==========================================
+ Coverage   85.37%   86.47%   +1.10%     
==========================================
  Files          77       92      +15     
  Lines        5660     6743    +1083     
==========================================
+ Hits         4832     5831     +999     
- Misses        828      912      +84     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

Metal Benchmarks

Details

Benchmark suite	Current: 162dfdf	Previous: 26c9e7d	Ratio
array/accumulate/Float32/1d	391000 ns	393834 ns	0.99
array/accumulate/Float32/dims=1	364292 ns	368959 ns	0.99
array/accumulate/Float32/dims=1L	8767416 ns	8770667 ns	1.00
array/accumulate/Float32/dims=2	432208 ns	429083 ns	1.01
array/accumulate/Float32/dims=2L	2530084 ns	2531958 ns	1.00
array/accumulate/Int64/1d	844166 ns	813000 ns	1.04
array/accumulate/Int64/dims=1	908125 ns	910333 ns	1.00
array/accumulate/Int64/dims=1L	9437250 ns	9446250 ns	1.00
array/accumulate/Int64/dims=2	1207833 ns	1210125 ns	1.00
array/accumulate/Int64/dims=2L	6424583 ns	6431042 ns	1.00
array/broadcast	239125 ns	238917 ns	1.00
array/construct	2333 ns	2292 ns	1.02
array/permutedims/2d	450541 ns	435125 ns	1.04
array/permutedims/3d	1032625 ns	997375 ns	1.04
array/permutedims/4d	1014625 ns	1092083 ns	0.93
array/private/copy	227000 ns	222750 ns	1.02
array/private/copyto!/cpu_to_gpu	213916 ns	212875 ns	1.00
array/private/copyto!/gpu_to_cpu	212209 ns	213375 ns	0.99
array/private/copyto!/gpu_to_gpu	215875 ns	211208 ns	1.02
array/private/iteration/findall/bool	1050000 ns	1027583 ns	1.02
array/private/iteration/findall/int	1213875 ns	1210334 ns	1.00
array/private/iteration/findfirst/bool	1148834 ns	1144291 ns	1.00
array/private/iteration/findfirst/int	1161125 ns	1151333 ns	1.01
array/private/iteration/findmin/1d	1205667 ns	1200334 ns	1.00
array/private/iteration/findmin/2d	1029625 ns	1023958 ns	1.01
array/private/iteration/logical	1653208 ns	1621875 ns	1.02
array/private/iteration/scalar	1388334 ns	1357334 ns	1.02
array/random/rand/Float32	415042 ns	418625 ns	0.99
array/random/rand/Int64	492833 ns	491042 ns	1.00
array/random/rand!/Float32	403584 ns	400584 ns	1.01
array/random/rand!/Int64	429958 ns	426750 ns	1.01
array/random/randn/Float32	385333 ns	387583 ns	0.99
array/random/randn!/Float32	373958 ns	372458 ns	1.00
array/reductions/mapreduce/Float32/1d	442625 ns	448792 ns	0.99
array/reductions/mapreduce/Float32/dims=1	300666 ns	348750 ns	0.86
array/reductions/mapreduce/Float32/dims=1L	608959 ns	612458 ns	0.99
array/reductions/mapreduce/Float32/dims=2	356458 ns	355000 ns	1.00
array/reductions/mapreduce/Float32/dims=2L	1074250 ns	1064708 ns	1.01
array/reductions/mapreduce/Int64/1d	628959 ns	626667 ns	1.00
array/reductions/mapreduce/Int64/dims=1	633667 ns	457375 ns	1.39
array/reductions/mapreduce/Int64/dims=1L	1011875 ns	1007000 ns	1.00
array/reductions/mapreduce/Int64/dims=2	763125 ns	791542 ns	0.96
array/reductions/mapreduce/Int64/dims=2L	2192041 ns	2208042 ns	0.99
array/reductions/reduce/Float32/1d	443167 ns	448000 ns	0.99
array/reductions/reduce/Float32/dims=1	345292 ns	357291 ns	0.97
array/reductions/reduce/Float32/dims=1L	615250 ns	616542 ns	1.00
array/reductions/reduce/Float32/dims=2	251833 ns	245792 ns	1.02
array/reductions/reduce/Float32/dims=2L	471834 ns	476459 ns	0.99
array/reductions/reduce/Int64/1d	631375 ns	632917 ns	1.00
array/reductions/reduce/Int64/dims=1	631542 ns	453375 ns	1.39
array/reductions/reduce/Int64/dims=1L	1005916 ns	1010625 ns	1.00
array/reductions/reduce/Int64/dims=2	262167 ns	264125 ns	0.99
array/reductions/reduce/Int64/dims=2L	668250 ns	671250 ns	1.00
array/shared/copy	130167 ns	130250 ns	1.00
array/shared/copyto!/cpu_to_gpu	38042 ns	37500 ns	1.01
array/shared/copyto!/gpu_to_cpu	36833 ns	38500 ns	0.96
array/shared/copyto!/gpu_to_gpu	37292 ns	38875 ns	0.96
array/shared/iteration/findall/bool	1055625 ns	1026708 ns	1.03
array/shared/iteration/findall/int	1221959 ns	1209666 ns	1.01
array/shared/iteration/findfirst/bool	924334 ns	966125 ns	0.96
array/shared/iteration/findfirst/int	987417 ns	979125 ns	1.01
array/shared/iteration/findmin/1d	1058625 ns	1064333 ns	0.99
array/shared/iteration/findmin/2d	1032667 ns	1030625 ns	1.00
array/shared/iteration/logical	1518708 ns	1376500 ns	1.10
array/shared/iteration/scalar	3864.5 ns	3744.75 ns	1.03
array/sorting/1d	2087167 ns	2081667 ns	1.00
array/sorting/2d	8296000 ns	8338750 ns	0.99
integration/byval/reference	1119125 ns	1123500 ns	1.00
integration/byval/slices=1	1121833 ns	1120292 ns	1.00
integration/byval/slices=2	2022334 ns	2027333 ns	1.00
integration/byval/slices=3	6635458 ns	6583833 ns	1.01
integration/metaldevrt	392750 ns	373250 ns	1.05
kernel/indexing	156625 ns	212750 ns	0.74
kernel/indexing_checked	398750 ns	391792 ns	1.02
kernel/launch	1820.8 ns	1804.2 ns	1.01
kernel/rand	398416 ns	404541 ns	0.98
latency/import	1789705375 ns	1786979292 ns	1.00
latency/precompile	32875539792 ns	32696140792 ns	1.01
latency/ttfp	2268129084 ns	2268758291 ns	1.00
metal/synchronization/context	557.5752688172043 ns	545.1957671957672 ns	1.02
metal/synchronization/stream	345.2916666666667 ns	350.5492957746479 ns	0.99

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