Merge pull request #331 from JuliaGPU/tb/int128

Don't rely on Int128

Author: maleadt

src/host/linalg.jl

@@ -110,16 +110,11 @@ function generic_matmatmul!(C::AnyGPUArray{R}, A::AnyGPUArray{T}, B::AnyGPUArray
         return fill!(C, zero(R))
     end
 
-    # reshape vectors to matrices
-    A′ = reshape(A, (size(A,1), size(A,2)))
-    B′ = reshape(B, (size(B,1), size(B,2)))
-    C′= reshape(C, (size(C,1), size(C,2)))
-
-    gpu_call(C′, A′, B′; name="matmatmul!") do ctx, C, A, B
+    gpu_call(C, A, B; name="matmatmul!") do ctx, C, A, B
         idx = @linearidx C
-        i, j = Tuple(CartesianIndices(C)[idx])
+        i, j = @inbounds Tuple(CartesianIndices(C)[idx])..., 1
 
-        if i <= size(A,1) && j <= size(B,2)
+        @inbounds if i <= size(A,1) && j <= size(B,2)
             z2 = zero(A[i, 1]*B[1, j] + A[i, 1]*B[1, j])
             Ctmp = convert(promote_type(R, typeof(z2)), z2)
             for k in 1:size(A,2)

test/testsuite/mapreduce.jl

@@ -70,19 +70,21 @@ end
                 @test compare(A->maximum(A; dims=dims), AT, rand(range, sz))
             end
         end
-        OT = isbitstype(widen(ET)) ? widen(ET) : ET
         for (sz,red) in [(10,)=>(1,), (10,10)=>(1,1), (10,10,10)=>(1,1,1), (10,10,10)=>(10,10,10),
                          (10,10,10)=>(1,10,10), (10,10,10)=>(10,1,10), (10,10,10)=>(10,10,1)]
             if !(ET <: Complex)
                 @test compare((A,R)->minimum!(R, A), AT, rand(range, sz), fill(typemax(ET), red))
                 @test compare((A,R)->maximum!(R, A), AT, rand(range, sz), fill(typemin(ET), red))
             end
         end
-        # smaller-scale test to avoid very large values and roundoff issues
-        for (sz,red) in [(2,)=>(1,), (2,2)=>(1,1), (2,2,2)=>(1,1,1), (2,2,2)=>(2,2,2),
-                         (2,2,2)=>(1,2,2), (2,2,2)=>(2,1,2), (2,2,2)=>(2,2,1)]
-            @test compare((A,R)->sum!(R, A), AT, rand(range, sz), rand(OT, red))
-            @test compare((A,R)->prod!(R, A), AT, rand(range, sz), rand(OT, red))
+        OT = isbitstype(widen(ET)) ? widen(ET) : ET
+        if OT in supported_eltypes()
+            # smaller-scale test to avoid very large values and roundoff issues
+            for (sz,red) in [(2,)=>(1,), (2,2)=>(1,1), (2,2,2)=>(1,1,1), (2,2,2)=>(2,2,2),
+                            (2,2,2)=>(1,2,2), (2,2,2)=>(2,1,2), (2,2,2)=>(2,2,1)]
+                @test compare((A,R)->sum!(R, A), AT, rand(range, sz), rand(OT, red))
+                @test compare((A,R)->prod!(R, A), AT, rand(range, sz), rand(OT, red))
+            end
         end
     end