[code-reflection] RFR: [hat] Type transform between bfloat16 <-> float improved

[Gary Frost]

On Tue, 9 Dec 2025 10:33:53 GMT, Juan Fumero <jfumero at openjdk.org> wrote:

> [hat] Type transform between bfloat16 <-> float improved.
> 
> Unittests with the OpenCL backend passing. 
> CUDA backend not passing due to another error (not related to this PR). We will merge this when we patch the CUDA backed.

hat/core/src/main/java/hat/codebuilders/C99HATCodeBuilder.java line 203:

> 201:     }
> 202: 
> 203:     public final T declareTypeBFloat16Type(String identifier) {

maybe  bf16Type()

hat/core/src/main/java/hat/codebuilders/C99HATKernelBuilder.java line 571:

> 569:         String identifier = "b";
> 570:         return funcDef(_ -> f32Type(),
> 571:                        _ -> builtin_bfloat16ToFloat(),

bf16Type...

hat/core/src/main/java/hat/codebuilders/C99HATKernelBuilder.java line 574:

> 572:                        _ -> ushortType(parameterName),
> 573:                        _ -> declareTypeBFloat16Type(identifier).semicolonNl()
> 574:                                .identifier(identifier).dot().identifier("s").osbrace().intConstZero().csbrace().equals().constant("0x0000").semicolonNl()

why constant("0x00000") why not intConstZer(0) or intvalue(0)

Replace osbrace() . xxx() .csbrace() === sbrace(_->xxx())

hat/core/src/main/java/hat/codebuilders/C99HATKernelBuilder.java line 594:

> 592:                 _->builtin_float2bfloat16(),
> 593:                 _->f32Type(parameterName),
> 594:                 _-> declareTypeBFloat16Type("b").space().equals().space().obrace().identifier(parameterName).cbrace().semicolonNl()

We can reduce the chances of misaligned braces by leaning into various helpers.      
   I think this becomes   something lik   
                   .assign(_->fb16Type("b"),_->brace(_->identifier(parameterName)).semicolonNl()
                   .returnKeyword(_->identifier("b").dot().identifier("s").sbrace(_->intConstOne()))

hat/core/src/main/java/hat/codebuilders/CodeBuilder.java line 588:

> 586: 
> 587:     public final T ushort() {
> 588:         return typeName("ushort");

u16Type  ??????

-------------

PR Review Comment: https://git.openjdk.org/babylon/pull/740#discussion_r2602242543
PR Review Comment: https://git.openjdk.org/babylon/pull/740#discussion_r2602247697
PR Review Comment: https://git.openjdk.org/babylon/pull/740#discussion_r2602255216
PR Review Comment: https://git.openjdk.org/babylon/pull/740#discussion_r2602225819
PR Review Comment: https://git.openjdk.org/babylon/pull/740#discussion_r2602234595