Simplify code generation for intrinsic fp16 mac, remove accidently co…

…mmited code
sklam · Dec 15, 2021 · Sep 21, 2021 · Sep 21, 2021 · Oct 5, 2021 · Oct 6, 2021
commit fd559aeb8e556f0fea585c466fb2fda74ca95012
diff --git a/numba/cuda/cudaimpl.py b/numba/cuda/cudaimpl.py
@@ -441,10 +441,8 @@ def ptx_hfma(context, builder, sig, args):
     hfma_inline_constraints = "=h,h,h,h"
     hfma_inline = ir.InlineAsm(hfma_fn_type, hfma_inline_asm,
                                hfma_inline_constraints)
-    arg1 = args[0]
-    arg2 = args[1]
-    arg3 = args[2]
-    return builder.call(hfma_inline, [arg1, arg2, arg3])
+
+    return builder.call(hfma_inline, args)
 
 # See:
 # https://docs.nvidia.com/cuda/libdevice-users-guide/__nv_cbrt.html#__nv_cbrt

diff --git a/numba/cuda/tests/cudadrv/test_managed_alloc.py b/numba/cuda/tests/cudadrv/test_managed_alloc.py
@@ -8,7 +8,6 @@
 
 
 @skip_on_cudasim('CUDA Driver API unsupported in the simulator')
-@unittest.skip
 @linux_only
 @skip_on_arm('Managed Alloc support is experimental/untested on ARM')
 class TestManagedAlloc(ContextResettingTestCase):