I should be sending up some AVX code this week. When I do this
I'd like to generate some testcases to make sure we actually
generate AVX code. Ideally we'd have a testcase for each AVX
pattern but that's probably overkill. Still, we'd like a lot
of tests, I think.
Should these tests go into CodeGen/X86 or should I create a new space,
like CodeGen/X86/SIMD? I tend to favor the latter as it keeps things
more compartmentalized and cleaner.
-Dave
On Jan 5, 2010, at 9:54 AM, David Greene wrote:> I should be sending up some AVX code this week. When I do this > I'd like to generate some testcases to make sure we actually > generate AVX code. Ideally we'd have a testcase for each AVX > pattern but that's probably overkill. Still, we'd like a lot > of tests, I think. > > Should these tests go into CodeGen/X86 or should I create a new space, > like CodeGen/X86/SIMD? I tend to favor the latter as it keeps things > more compartmentalized and cleaner.Adding a subdirectory here sounds good. "SIMD" might be a bit too general though; how about test/CodeGen/X86/AVX? Dan
On Tuesday 05 January 2010 13:20, Dan Gohman wrote:> On Jan 5, 2010, at 9:54 AM, David Greene wrote: > > I should be sending up some AVX code this week. When I do this > > I'd like to generate some testcases to make sure we actually > > generate AVX code. Ideally we'd have a testcase for each AVX > > pattern but that's probably overkill. Still, we'd like a lot > > of tests, I think. > > > > Should these tests go into CodeGen/X86 or should I create a new space, > > like CodeGen/X86/SIMD? I tend to favor the latter as it keeps things > > more compartmentalized and cleaner. > > Adding a subdirectory here sounds good. "SIMD" might be a bit too general > though; how about test/CodeGen/X86/AVX?This will be a rewrite of all the x86 SIMD stuff so the name seemed appropriate. It'll be testing the whole infrastructure, not just AVX. -Dave