device-libs: Remove clamping for f16 ldexp values

arsenm · arsenm · commit 8fc01a7bcdf7 · 2025-04-19T09:57:44.000+02:00
The generic builtin handles this correctly.

Change-Id: I12cb2ed1c5f4661b4b29df79987a872a17a1a43a
diff --git a/amd/device-libs/ocml/src/ldexpH.cl b/amd/device-libs/ocml/src/ldexpH.cl
@@ -16,6 +16,6 @@ MATH_MANGLE2(ldexp)(half2 x, int2 n)
 CONSTATTR half
 MATH_MANGLE(ldexp)(half x, int n)
 {
-    return BUILTIN_FLDEXP_F16(x, BUILTIN_CLAMP_S32(n, SHRT_MIN, SHRT_MAX));
+    return BUILTIN_FLDEXP_F16(x, n);
 }
 

Original file line number	Diff line number	Diff line change
`@@ -16,6 +16,6 @@ MATH_MANGLE2(ldexp)(half2 x, int2 n)`
`16`	`16`	`CONSTATTR half`
`17`	`17`	`MATH_MANGLE(ldexp)(half x, int n)`
`18`	`18`	`{`
`19`		`- return BUILTIN_FLDEXP_F16(x, BUILTIN_CLAMP_S32(n, SHRT_MIN, SHRT_MAX));`
	`19`	`+ return BUILTIN_FLDEXP_F16(x, n);`
`20`	`20`	`}`
`21`	`21`