Adding IsFinite, IsInf and IsNan long vector tests #7836

joaosaffran · 2025-10-24T20:00:18Z

This PR adds isfinite, isinf, isnan intrinsic test to long vectors.

Closes: #7850

github-actions · 2025-10-27T18:31:26Z

✅ With the latest revision this PR passed the C/C++ code formatter.

alsepkow · 2025-10-28T01:27:16Z

tools/clang/unittests/HLSLExec/LongVectorTestData.h

 INPUT_SET(InputSet::Positive, 1.0f, 1.0f, 65535.0f, 0.01f, 5531.0f, 0.01f, 1.0f,
          0.01f, 331.2330f, 3250.01f);
 INPUT_SET(InputSet::SelectCond, 0.0f, 1.0f);
+INPUT_SET(InputSet::Infinite, std::numeric_limits<float>::infinity(),


You'll want to add some finite values to the set of values we test. Otherwise, all the results are just going to be false.

Looking forward a bit, we know that we're going to be adding some more special values? Would it be better to have an input set that contains a mixture of non-special and special float values?

alsepkow · 2025-10-28T01:34:06Z

tools/clang/unittests/HLSLExec/LongVectorTestData.h

 INPUT_SET(InputSet::Positive, 1.0, 1.0, 342.0, 0.01, 5531.0, 0.01, 1.0, 0.01,
          331.2330, 3250.01);
 INPUT_SET(InputSet::SelectCond, 0.0, 1.0);
+INPUT_SET(InputSet::Infinite, std::numeric_limits<HLSLHalf_t>::infinity(),


You should double check what's happening here. Passing HLSLHalf_t as the template argument to std::numeric_limits will be doing an implicit conversion to float. Which is fine. And it looks like DirectX::PackedVector::XMConvertFloatToHalf does the right thing by converting that to an inf 'HALF'.

I think it would be better to do std::numeric_limits::infinity() to be more explicit. What do you think?

alsepkow · 2025-10-28T01:34:46Z

tools/clang/unittests/HLSLExec/LongVectors.cpp

+// Float Ops
+//
+
+#define BOOLEAN_FLOAT_OP(OP, IMPL)                                             \


I'm assuming this macro will be helpful for the other float ops we're adding.

"boolean float op" sounds a bit contradictory.

I think we're doing a bunch of "float specials" here, I wonder if it would make sense to group them in that way?

alsepkow

Some minor suggestions. Outside of those LGTM

damyanp · 2025-10-28T16:10:18Z

Please associate this with the appropriate workitem (maybe with a "closes #xxx" type thing in the PR description)

damyanp · 2025-10-28T16:19:07Z

tools/clang/unittests/HLSLExec/LongVectorOps.def

 OP(Ternary, Select, 3, "TestSelect", "", " -DFUNC_TEST_SELECT=1",
   "LongVectorOp", SelectCond, Default2, Default3)

+OP(Unary, IsFinite, 1, "TestIsFinite", "", " -DFUNC_TEST_ISFINITE=1",


Is Unary the right group to use here? Why not UnaryMath (and if so, should move this up next to the other unary math ops)

I suggested to use 'FloatSpecial' as I don't think this is really a 'Math' Op either. 'UnaryLogical' would also fit IMO. But 'FloatSpecial' seems even better.

(no need to change anything, calling it FloatSpecial is fine, but I can't resist responding....)

I don't think this is really a 'Math' Op either

The C implementations of these are all provided by math.h, and the C++ ones are in cmath....so, they're usually grouped in as math functions.

Poor choice of wording on my part. It wasn't that it's not a math op and more so that Math op felt too generic for this context. The trig functions are technically unary math ops as well but grouping them as trig ops felt logical.

Just sharing my thoughts there!

damyanp · 2025-10-28T16:21:27Z

tools/clang/unittests/HLSLExec/LongVectorTestData.h

 INPUT_SET(InputSet::Positive, 1.0f, 1.0f, 65535.0f, 0.01f, 5531.0f, 0.01f, 1.0f,
          0.01f, 331.2330f, 3250.01f);
 INPUT_SET(InputSet::SelectCond, 0.0f, 1.0f);
+INPUT_SET(InputSet::Infinite, std::numeric_limits<float>::infinity(),


Looking forward a bit, we know that we're going to be adding some more special values? Would it be better to have an input set that contains a mixture of non-special and special float values?

damyanp · 2025-10-28T16:23:40Z

tools/clang/unittests/HLSLExec/LongVectors.cpp

+// Float Ops
+//
+
+#define BOOLEAN_FLOAT_OP(OP, IMPL)                                             \


"boolean float op" sounds a bit contradictory.

I think we're doing a bunch of "float specials" here, I wonder if it would make sense to group them in that way?

damyanp · 2025-10-28T16:23:56Z

tools/clang/unittests/HLSLExec/LongVectors.cpp

+  HLK_TEST(IsFinite, double);
  // Explicit Cast


Suggested change

HLK_TEST(IsFinite, double);

// Explicit Cast

HLK_TEST(IsFinite, double);

// Explicit Cast

damyanp · 2025-10-28T16:24:59Z

tools/clang/unittests/HLSLExec/ShaderOpArith.xml

        }
        #endif

+        #ifdef FUNC_TEST_ISFINITE


Should this really be something like "FLOAT_SPECIALS" and have another define choose between isinifite, isinf and isnan?

I was checking LongVectorOps.def, no other test requires 2 defines to specify the test. In some cases, like IS_REDUCTION_OP it modifies other parts of the code instead of defining functions to test. So I am not really sure if we should modify this for those specific tests. Unless the FLOAT_SPECIALS define modify some other parts of the code.

Aren't these IS_UNARY_OP with FUNC being one of isinfinite, isinf and isnan, and OUT_TYPE being bool?

You don't actually need to define new helper functions in the HLSL source for these cases.
The 4th argument to the 'OP' macro in LongVectorOps.def is 'INTRINSIC'. Currently you're setting it to 'TestIsFinite', 'TestIsInf', or 'TestIsNan'.

Instead, you can put the actual HLSL intrinsic name there and the generic path in this test xml will handle it for you!
The only other thing to worry about with that is setting OUT_TYPE to bool, which you're already doing implicitly via the return type via the operator overload in your FLOAT_SPECIAL_OP macro :).

TLDR; Get rid of these function definitions and update the string in the 'OP' macros in LongVectorOps.def to be the actual intrinsic name.

Edit: You can actually swap to using the 'OP_DEFAULT' macro because you wont need custom defines anymore.

tools/clang/unittests/HLSLExec/LongVectors.cpp

damyanp · 2025-10-29T21:05:39Z

tools/clang/unittests/HLSLExec/ShaderOpArith.xml

        }
        #endif

+        #ifdef FUNC_TEST_ISFINITE


Aren't these IS_UNARY_OP with FUNC being one of isinfinite, isinf and isnan, and OUT_TYPE being bool?

alsepkow · 2025-10-29T22:03:43Z

tools/clang/unittests/HLSLExec/LongVectorOps.def

 OP_DEFAULT(UnaryMath, Log2, 1, "log2", "")
 OP_DEFAULT_DEFINES(UnaryMath, Frexp, 1, "TestFrexp", "", " -DFUNC_FREXP=1")

+OP(UnaryMath, IsFinite, 1, "TestIsFinite", "", " -DFUNC_TEST_ISFINITE=1",


nit: These aren't really 'UnaryMath' operations. The name isn't currently used for anything outside of a human hint.
But I'm thinking 'FloatSpecial' would be better. My main nit is that these are logical operations, not math operations.

alsepkow · 2025-10-29T22:05:25Z

tools/clang/unittests/HLSLExec/LongVectors.cpp

  HLK_TEST(Frexp, float);
  HLK_TEST(Abs, double);
  HLK_TEST(Sign, double);



Suggested change

// Float Special

alsepkow · 2025-10-29T22:07:08Z

tools/clang/unittests/HLSLExec/LongVectors.cpp

+  HLK_TEST(IsInf, float);
+
+  HLK_TEST(IsNan, HLSLHalf_t);
+  HLK_TEST(IsNan, float);


nit: The other test cases are organized in groups by input type

Suggested change

HLK_TEST(IsNan, float);

HLK_TEST(IsInf, HLSLHalf_t);

HLK_TEST(IsFinite, HLSLHalf_t);

HLK_TEST(IsNan, HLSLHalf_t);

HLK_TEST(IsInf, float);

HLK_TEST(IsFinite, float);

HLK_TEST(IsNan, float);

alsepkow · 2025-10-29T22:19:06Z

tools/clang/unittests/HLSLExec/LongVectorTestData.h

 INPUT_SET(InputSet::Positive, 1.0, 1.0, 342.0, 0.01, 5531.0, 0.01, 1.0, 0.01,
          331.2330, 3250.01);
 INPUT_SET(InputSet::SelectCond, 0.0, 1.0);
+// HLSLHalf_t cast this to float when assigning it.


Suggested change

// HLSLHalf_t cast this to float when assigning it.

// HLSLHalf_t has a constructor which accepts a float and converts it to half precision by

// clamping to the representable range via DirectX::PackedVector::XMConvertFloatToHalf.

alsepkow

LGTM. Thanks for your help!

adding test?

f54718e

github-project-automation bot added this to HLSL Roadmap Oct 24, 2025

github-project-automation bot moved this to New in HLSL Roadmap Oct 24, 2025

joaosaffran added 4 commits October 24, 2025 13:02

remove input set

b8585b8

adding test

0495fd5

fix

271cb0a

Merge branch 'main' into test/isfinite

20eaa93

joaosaffran added 4 commits October 27, 2025 12:08

adding other numeric types test

27f4e31

clean up

5d66ed8

reduce tests to only float point types

068f4fc

format

c51afa9

joaosaffran marked this pull request as ready for review October 27, 2025 19:16

joaosaffran requested a review from alsepkow October 27, 2025 19:21

alsepkow reviewed Oct 28, 2025

View reviewed changes

joaosaffran linked an issue Oct 28, 2025 that may be closed by this pull request

Long Vector Execution Tests: Float Specific Ops #7694

Closed

damyanp reviewed Oct 28, 2025

View reviewed changes

joaosaffran marked this pull request as draft October 28, 2025 17:37

joaosaffran-zz and others added 3 commits October 28, 2025 17:57

add weird values

dbbc868

remove double

eab4ea4

adding other tests

a4de02c

joaosaffran marked this pull request as ready for review October 29, 2025 20:12

joaosaffran changed the title ~~Adding IsFinite long vector tests~~ Adding IsFinite, IsInf and IsNan long vector tests Oct 29, 2025

joaosaffran requested review from alsepkow and damyanp October 29, 2025 20:13

rename input set

57a6be9

damyanp reviewed Oct 29, 2025

View reviewed changes

alsepkow reviewed Oct 29, 2025

View reviewed changes

joaosaffran added 3 commits October 29, 2025 15:54

using OP_DEFAULT

76ca7ce

addressing comments

3bd8288

format

500ba14

joaosaffran requested review from alsepkow and damyanp October 29, 2025 23:05

alsepkow approved these changes Oct 29, 2025

View reviewed changes

format

56930db

damyanp approved these changes Oct 29, 2025

View reviewed changes

joaosaffran merged commit b3619bf into microsoft:main Oct 30, 2025
12 checks passed

github-project-automation bot moved this from New to Done in HLSL Roadmap Oct 30, 2025

-// HLSLHalf_t cast this to float when assigning it.
+// HLSLHalf_t has a constructor which accepts a float and converts it to half precision by
+// clamping to the representable range via DirectX::PackedVector::XMConvertFloatToHalf.

Adding IsFinite, IsInf and IsNan long vector tests #7836

Adding IsFinite, IsInf and IsNan long vector tests #7836

Conversation

joaosaffran commented Oct 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Oct 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alsepkow left a comment

Choose a reason for hiding this comment

Uh oh!

damyanp commented Oct 28, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alsepkow Oct 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alsepkow Oct 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alsepkow left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

joaosaffran commented Oct 24, 2025 •

edited

Loading

github-actions bot commented Oct 27, 2025 •

edited

Loading

alsepkow Oct 29, 2025 •

edited

Loading

alsepkow Oct 29, 2025 •

edited

Loading