Fix the LargestVectorSize parameter for SVE test cases #100606
Conversation
|
@dotnet/arm64-contrib |
|
Tagging subscribers to this area: @dotnet/area-system-runtime-intrinsics |
| ("SveLoadMaskedUnOpTest.template", new Dictionary<string, string> { ["TestName"] = "SveLoadVector_ushort", ["Isa"] = "Sve", ["Method"] = "LoadVector", ["RetVectorType"] = "Vector", ["RetBaseType"] = "UInt16", ["Op1VectorType"] = "Vector", ["Op1BaseType"] = "UInt16", ["Op2BaseType"] = "UInt16", ["LargestVectorSize"] = "8", ["NextValueOp2"] = "TestLibrary.Generator.GetUInt16()", ["ValidateIterResult"] = "firstOp[i] != result[i]"}), | ||
| ("SveLoadMaskedUnOpTest.template", new Dictionary<string, string> { ["TestName"] = "SveLoadVector_uint", ["Isa"] = "Sve", ["Method"] = "LoadVector", ["RetVectorType"] = "Vector", ["RetBaseType"] = "UInt32", ["Op1VectorType"] = "Vector", ["Op1BaseType"] = "UInt32", ["Op2BaseType"] = "UInt32", ["LargestVectorSize"] = "8", ["NextValueOp2"] = "TestLibrary.Generator.GetUInt32()", ["ValidateIterResult"] = "firstOp[i] != result[i]"}), | ||
| ("SveLoadMaskedUnOpTest.template", new Dictionary<string, string> { ["TestName"] = "SveLoadVector_ulong", ["Isa"] = "Sve", ["Method"] = "LoadVector", ["RetVectorType"] = "Vector", ["RetBaseType"] = "UInt64", ["Op1VectorType"] = "Vector", ["Op1BaseType"] = "UInt64", ["Op2BaseType"] = "UInt64", ["LargestVectorSize"] = "8", ["NextValueOp2"] = "TestLibrary.Generator.GetUInt64()", ["ValidateIterResult"] = "firstOp[i] != result[i]"}), | ||
| ("SveLoadMaskedUnOpTest.template", new Dictionary<string, string> { ["TestName"] = "SveLoadVector_float", ["Isa"] = "Sve", ["Method"] = "LoadVector", ["RetVectorType"] = "Vector", ["RetBaseType"] = "Single", ["Op1VectorType"] = "Vector", ["Op1BaseType"] = "Single", ["Op2BaseType"] = "Single", ["LargestVectorSize"] = "16", ["NextValueOp2"] = "TestLibrary.Generator.GetSingle()", ["ValidateIterResult"] = "firstOp[i] != result[i]"}), |
There was a problem hiding this comment.
This would fail if the tests were run on say the AWS Graviton, where LargestVectorSize == 32 (256-bit vectors).
The actual limit for SVE vector size is 2048 bits and so we should probably set LargestVectorSize = 256 (although the highest shipped hardware to date is 512-bits, or rather LargestVectorSize = 64) to ensure we're future proof.
There was a problem hiding this comment.
Is LargestVectorSize only used to ensure it creates a large enough pool of memory for the data values?
There was a problem hiding this comment.
Yes, it's used to ensure that there is enough data to load from memory and to store to memory so that it can be tested/validated.
To that extent, it needs to be at least as big as any hardware we run on supports; but more ideally should just be future proof as well.
|
i will fix it as part of #100743 |
Fixes: #100366