[maleadt]

The relocation patching test read a libjulia global through GPUCompiler.Runtime.type_tag(Val(:float32)). JuliaGPU/GPUCompiler.jl#1067 removes that method, since the runtime no longer loads type tags from data symbols. Emit the load from jl_float32_type with llvmcall instead, which exercises the same relocation without depending on GPUCompiler internals.

[github-actions]

Benchmark results

Compared with d32eb2641, the merge base.
The environment differs from the baseline's: GPUArrays 12.1.1 → 12.2.0, GPUArraysCore 0.2.1 → 0.2.2, SPIRVIntrinsics 1.5.1 → 1.6.0.

Times are the minimum time of every benchmark. A change is unusual when it exceeds both three times the benchmark's typical change between consecutive runs on the default branch and a minimum per category (performance 3%, latency 5%, applications 3%, other 3%). A benchmark that starts or stops allocating is listed however few the allocations. A single run can still be a fluke, so rerun before drawing conclusions.

	Geometric mean	Unusual: slower	Unusual: faster	Compared
Performance	+0.1%	0	0	79
Latency	+0.2%	0	0	3
Applications	−0.1%	0	0	6

No unusual changes.

All benchmarks

Benchmark	Baseline	This PR	Change
Performance
array/accumulate/Float32/1d	39.5 µs	40.0 µs	+1.4%
array/accumulate/Float32/dims=1	29.1 µs	29.7 µs	+1.9%
array/accumulate/Float32/dims=1L	83.6 µs	83.8 µs	+0.3%
array/accumulate/Float32/dims=2	48.1 µs	48.6 µs	+0.9%
array/accumulate/Float32/dims=2L	372 µs	372 µs	−0.0%
array/accumulate/Int64/1d	70.8 µs	71.4 µs	+0.9%
array/accumulate/Int64/dims=1	46.6 µs	46.3 µs	−0.6%
array/accumulate/Int64/dims=1L	142 µs	142 µs	+0.0%
array/accumulate/Int64/dims=2	68.0 µs	68.3 µs	+0.4%
array/accumulate/Int64/dims=2L	765 µs	768 µs	+0.4%
array/broadcast	9.50 µs	9.42 µs	−0.9%
array/broadcast launch	5.89 µs	5.78 µs	−1.7%
array/construct	950 ns	917 ns	−3.4%
array/copy	17.2 µs	17.4 µs	+1.1%
array/copyto!/cpu_to_gpu	208 µs	210 µs	+1.2%
array/copyto!/gpu_to_cpu	241 µs	242 µs	+0.7%
array/copyto!/gpu_to_gpu	8.82 µs	8.95 µs	+1.5%
array/iteration/findall/bool	46.0 µs	46.5 µs	+1.0%
array/iteration/findall/int	59.3 µs	60.8 µs	+2.6%
array/iteration/findfirst/bool	33.6 µs	33.8 µs	+0.6%
array/iteration/findfirst/int	33.6 µs	33.7 µs	+0.2%
array/iteration/findmin/1d	51.2 µs	52.0 µs	+1.6%
array/iteration/findmin/2d	47.1 µs	48.1 µs	+2.0%
array/iteration/logical	45.9 µs	46.8 µs	+2.1%
array/iteration/scalar	60.1 µs	60.0 µs	−0.2%
array/permutedims/2d	46.1 µs	46.0 µs	−0.2%
array/permutedims/3d	43.5 µs	43.1 µs	−1.0%
array/permutedims/4d	36.9 µs	36.0 µs	−2.2%
array/random/rand!/Float32	8.36 µs	8.48 µs	+1.4%
array/random/rand!/Int64	14.7 µs	14.6 µs	−0.6%
array/random/rand/Float32	11.4 µs	11.3 µs	−1.6%
array/random/rand/Int64	16.8 µs	16.7 µs	−0.9%
array/random/randn!/Float32	25.1 µs	25.2 µs	+0.7%
array/random/randn/Float32	33.8 µs	33.6 µs	−0.6%
array/reductions/mapreduce/Float32/1d	31.6 µs	32.0 µs	+1.4%
array/reductions/mapreduce/Float32/dims=1	17.2 µs	17.0 µs	−1.2%
array/reductions/mapreduce/Float32/dims=1L	54.5 µs	54.5 µs	−0.0%
array/reductions/mapreduce/Float32/dims=2	24.0 µs	24.2 µs	+0.9%
array/reductions/mapreduce/Float32/dims=2L	55.1 µs	54.6 µs	−0.8%
array/reductions/mapreduce/Int64/1d	31.5 µs	31.8 µs	+0.9%
array/reductions/mapreduce/Int64/dims=1	18.9 µs	19.0 µs	+0.7%
array/reductions/mapreduce/Int64/dims=1L	93.4 µs	93.5 µs	+0.1%
array/reductions/mapreduce/Int64/dims=2	25.4 µs	25.4 µs	−0.3%
array/reductions/mapreduce/Int64/dims=2L	79.8 µs	80.6 µs	+0.9%
array/reductions/reduce/Float32/1d	31.3 µs	32.0 µs	+2.1%
array/reductions/reduce/Float32/dims=1	17.0 µs	16.9 µs	−0.6%
array/reductions/reduce/Float32/dims=1L	54.6 µs	54.7 µs	+0.3%
array/reductions/reduce/Float32/dims=2	24.0 µs	23.7 µs	−1.5%
array/reductions/reduce/Float32/dims=2L	55.0 µs	54.7 µs	−0.6%
array/reductions/reduce/Int64/1d	31.6 µs	31.6 µs	−0.0%
array/reductions/reduce/Int64/dims=1	18.9 µs	18.9 µs	−0.3%
array/reductions/reduce/Int64/dims=1L	93.6 µs	93.8 µs	+0.1%
array/reductions/reduce/Int64/dims=2	25.3 µs	25.5 µs	+0.8%
array/reductions/reduce/Int64/dims=2L	80.4 µs	80.9 µs	+0.7%
array/reverse/1d	18.6 µs	18.9 µs	+1.5%
array/reverse/1d_inplace	11.5 µs	11.3 µs	−1.7%
array/reverse/1dL	72.7 µs	72.6 µs	−0.1%
array/reverse/1dL_inplace	69.8 µs	69.4 µs	−0.5%
array/reverse/2d	21.4 µs	21.3 µs	−0.2%
array/reverse/2d_inplace	13.9 µs	13.9 µs	−0.2%
array/reverse/2dL	79.8 µs	79.9 µs	+0.1%
array/reverse/2dL_inplace	70.1 µs	69.9 µs	−0.3%
array/sorting/1d	651 µs	651 µs	−0.1%
array/sorting/2d	177 µs	177 µs	−0.2%
array/sorting/by	851 µs	853 µs	+0.3%
cuda/graph/capture	27.7 µs	27.7 µs	+0.0%
cuda/graph/eager	27.8 µs	25.3 µs	−9.1%
cuda/graph/launch	14.0 µs	14.0 µs	−0.1%
cuda/synchronization/context/auto	2.35 µs	2.38 µs	+1.1%
cuda/synchronization/context/blocking	933 ns	913 ns	−2.1%
cuda/synchronization/context/nonblocking	10.7 µs	11.1 µs	+3.7%
cuda/synchronization/stream/auto	886 ns	875 ns	−1.2%
cuda/synchronization/stream/blocking	776 ns	780 ns	+0.6%
cuda/synchronization/stream/nonblocking	8.34 µs	8.67 µs	+3.9%
kernel/indexing	10.5 µs	10.5 µs	+0.4%
kernel/indexing_checked	11.2 µs	11.2 µs	+0.1%
kernel/launch	1.38 µs	1.39 µs	+1.4%
kernel/occupancy	34.2 ns	34.2 ns	−0.0%
kernel/rand	14.0 µs	14.0 µs	−0.3%
Latency
import	3.95 s	3.95 s	+0.1%
precompile	4.91 s	4.92 s	+0.2%
ttfp	4.46 s	4.49 s	+0.5%
Applications
byval/reference	147 µs	147 µs	−0.2%
byval/slices=1	148 µs	147 µs	−0.2%
byval/slices=2	289 µs	289 µs	−0.0%
byval/slices=3	431 µs	431 µs	−0.1%
cudadevrt	104 µs	104 µs	−0.1%
volumerhs	9.61 ms	9.61 ms	−0.0%

History: https://juliagpu.org/perf/?pkg=cuda

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 86.20%. Comparing base (d32eb26) to head (a98f7fe).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #3500       +/-   ##
===========================================
+ Coverage   43.88%   86.20%   +42.32%     
===========================================
  Files         196      196               
  Lines       21356    21247      -109     
===========================================
+ Hits         9372    18317     +8945     
+ Misses      11984     2930     -9054     

☔ 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.