[maleadt]

Base compares Rationals with each other, with integers and with floats by multiplying out their denominators with widemul. With 64-bit rationals (or any float, whose decompose returns Ints) that computes Int128 products, which SPIR-V doesn't support. This stacks GPUToolbox's new int128_overrides table (JuliaGPU/GPUToolbox.jl#30) regardless of fp64 support, and requires GPUToolbox 3.4. Same fix as JuliaGPU/Metal.jl#550.

Tested on an Iris Xe: the new tests fail without the change (13 errors) and pass with it.

[github-actions]

Your PR requires formatting changes to meet the project's style guidelines.
Please consider running Runic (git runic main) to apply these changes.

Click here to view the suggested changes.

diff --git a/src/compiler/compilation.jl b/src/compiler/compilation.jl
index 2bfff9c..5c56740 100644
--- a/src/compiler/compilation.jl
+++ b/src/compiler/compilation.jl
@@ -42,8 +42,10 @@ function GPUCompiler.method_tables(job::oneAPICompilerJob)
         (method_table, SPIRVIntrinsics.method_table, GPUToolbox.Overlays.int128_overrides)
     else
         # keep single-precision math that Base computes in Float64 out of double precision
-        (method_table, SPIRVIntrinsics.method_table, GPUToolbox.Overlays.float64_overrides,
-         GPUToolbox.Overlays.int128_overrides)
+        (
+            method_table, SPIRVIntrinsics.method_table, GPUToolbox.Overlays.float64_overrides,
+            GPUToolbox.Overlays.int128_overrides,
+        )
     end
 end
 
diff --git a/test/device/intrinsics.jl b/test/device/intrinsics.jl
index ce37747..eeab908 100644
--- a/test/device/intrinsics.jl
+++ b/test/device/intrinsics.jl
@@ -54,9 +54,9 @@ end
         xs = Float32[0, -1, 1.4, 1.5, 1.6, Inf, NaN]
         @test Array(map(x -> x < r, oneArray(xs))) == (xs .< r)
 
-        qs = [1//3, -5//7, 3//2, typemax(Int)//2, 2^40//3, -typemax(Int)//(2^40 + 1)]
+        qs = [1 // 3, -5 // 7, 3 // 2, typemax(Int) // 2, 2^40 // 3, -typemax(Int) // (2^40 + 1)]
         ns = [0, -1, 2, 2^35, typemax(Int), typemin(Int)]
-        fs = Float32[0.3, -0.7, 1.5, 2f0^62, 3.6f11, -Inf]
+        fs = Float32[0.3, -0.7, 1.5, 2.0f0^62, 3.6f11, -Inf]
         for op in (<, <=, cmp)
             @test Array(op.(oneArray(qs), oneArray(fs))) == op.(qs, fs)
             @test Array(op.(oneArray(fs), oneArray(qs))) == op.(fs, qs)

[maleadt]

CI failure is a bug in NEO. Claude has a fix.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.05%. Comparing base (8460d3f) to head (aa6f3a3).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #707      +/-   ##
==========================================
- Coverage   83.70%   83.05%   -0.66%     
==========================================
  Files          52       51       -1     
  Lines        4376     4254     -122     
==========================================
- Hits         3663     3533     -130     
- Misses        713      721       +8     

☔ 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.