[libclc] [libclc] Refine __clc_fp*_subnormals_supported and __clc_flush_denormal_if_not_supported (PR #157633)

[Matt Arsenault via cfe-commits]

================
@@ -66,13 +65,11 @@ bool __attribute__((noinline)) __clc_runtime_has_hw_fma32(void);
 #define LOG_MAGIC_NUM_SP32 (1 + NUMEXPBITS_SP32 - EXPBIAS_SP32)
 
 _CLC_OVERLOAD _CLC_INLINE float __clc_flush_denormal_if_not_supported(float x) {
-  int ix = __clc_as_int(x);
-  if (!__clc_fp32_subnormals_supported() && ((ix & EXPBITS_SP32) == 0) &&
-      ((ix & MANTBITS_SP32) != 0)) {
-    ix &= SIGNBIT_SP32;
-    x = __clc_as_float(ix);
-  }
-  return x;
+  // Avoid calling __clc_fp32_subnormals_supported here: it uses
+  // llvm.canonicalize, which quiets sNaN.
+  return __builtin_fabsf(x) < 0x1p-149f
+             ? __builtin_elementwise_copysign(0.0f, x)
+             : x;
----------------
arsenm wrote:

The name is now right, but the usage is not. You do not want to unconditionally force a denormal flush, pretty much anywhere. 

I also believe most of the explicit flushes / canonicalizes in the math function implementations can be deleted with some shuffling around. I removed most of these in the rocm-device-libs implementations a few years ago. A few more remain but I believe they can also be avoided 

https://github.com/llvm/llvm-project/pull/157633