[libclc] Move fmin & fmax to CLC library (llvm#134218)

This is an alternative to llvm#128506 which doesn't attempt to change the
codegen for fmin and fmax on their way to the CLC library.

The amdgcn and r600 custom definitions of fmin/fmax are now converted to
custom definitions of __clc_fmin and __clc_fmax.

For simplicity, the CLC library doesn't provide vector/scalar versions
of these builtins. The OpenCL layer wraps those up to the vector/vector
versions.

The only codegen change is that non-standard vector/scalar overloads of
fmin/fmax have been removed. We were currently (accidentally,
presumably) providing overloads with mixed elment types such as
fmin(double2, float), fmax(half4, double), etc. The only vector/scalar
overloads in the OpenCL spec are those with scalars of the same element
type as the vector in the first argument.

Author: frasercrmck

libclc/amdgcn/lib/SOURCES

@@ -1,6 +1,4 @@
 cl_khr_int64_extended_atomics/minmax_helpers.ll
-math/fmax.cl
-math/fmin.cl
 mem_fence/fence.cl
 synchronization/barrier.cl
 workitem/get_global_offset.cl

libclc/amdgcn/lib/math/fmax.cl

@@ -0,0 +1,20 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __CLC_MATH_CLC_FMAX_H__
+#define __CLC_MATH_CLC_FMAX_H__
+
+#define __CLC_FUNCTION __clc_fmax
+#define __CLC_BODY <clc/shared/binary_decl.inc>
+
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_FMAX_H__

libclc/amdgcn/lib/math/fmin.cl

@@ -0,0 +1,20 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __CLC_MATH_CLC_FMIN_H__
+#define __CLC_MATH_CLC_FMIN_H__
+
+#define __CLC_FUNCTION __clc_fmin
+#define __CLC_BODY <clc/shared/binary_decl.inc>
+
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_FMIN_H__

libclc/clc/include/clc/math/clc_fmax.h

@@ -0,0 +1,15 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION(__CLC_GENTYPE x,
+                                                     __CLC_GENTYPE y);
+
+#ifndef __CLC_SCALAR
+_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION(__CLC_GENTYPE x,
+                                                     __CLC_SCALAR_GENTYPE y);
+#endif

libclc/clc/include/clc/math/clc_fmin.h

@@ -0,0 +1,25 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/utils.h>
+
+#ifndef __CLC_FUNCTION
+#define __CLC_FUNCTION(x) __CLC_CONCAT(__clc_, x)
+#endif
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE FUNCTION(__CLC_GENTYPE a,
+                                              __CLC_GENTYPE b) {
+  return __CLC_FUNCTION(FUNCTION)(a, b);
+}
+
+#ifndef __CLC_SCALAR
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE FUNCTION(__CLC_GENTYPE a,
+                                              __CLC_SCALAR_GENTYPE b) {
+  return __CLC_FUNCTION(FUNCTION)(a, (__CLC_GENTYPE)b);
+}
+#endif

libclc/clc/include/clc/shared/binary_decl_with_scalar_second_arg.inc

@@ -1 +1,3 @@
+math/clc_fmax.cl
+math/clc_fmin.cl
 math/clc_ldexp_override.cl

libclc/clc/include/clc/shared/binary_def_with_scalar_second_arg.inc

@@ -0,0 +1,49 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/clcmacro.h>
+#include <clc/internal/clc.h>
+#include <clc/relational/clc_isnan.h>
+
+_CLC_DEF _CLC_OVERLOAD float __clc_fmax(float x, float y) {
+  // fcanonicalize removes sNaNs and flushes denormals if not enabled. Otherwise
+  // fmax instruction flushes the values for comparison, but outputs original
+  // denormal
+  x = __builtin_canonicalizef(x);
+  y = __builtin_canonicalizef(y);
+  return __builtin_fmaxf(x, y);
+}
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, float, __clc_fmax, float, float)
+
+#ifdef cl_khr_fp64
+
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+
+_CLC_DEF _CLC_OVERLOAD double __clc_fmax(double x, double y) {
+  x = __builtin_canonicalize(x);
+  y = __builtin_canonicalize(y);
+  return __builtin_fmax(x, y);
+}
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, double, __clc_fmax, double,
+                      double)
+
+#endif
+#ifdef cl_khr_fp16
+
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+
+_CLC_DEF _CLC_OVERLOAD half __clc_fmax(half x, half y) {
+  if (__clc_isnan(x))
+    return y;
+  if (__clc_isnan(y))
+    return x;
+  return (y < x) ? x : y;
+}
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, half, __clc_fmax, half, half)
+
+#endif

libclc/clc/lib/amdgcn/SOURCES

@@ -0,0 +1,50 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/clcmacro.h>
+#include <clc/internal/clc.h>
+#include <clc/relational/clc_isnan.h>
+
+_CLC_DEF _CLC_OVERLOAD float __clc_fmin(float x, float y) {
+  // fcanonicalize removes sNaNs and flushes denormals if not enabled. Otherwise
+  // fmin instruction flushes the values for comparison, but outputs original
+  // denormal
+  x = __builtin_canonicalizef(x);
+  y = __builtin_canonicalizef(y);
+  return __builtin_fminf(x, y);
+}
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, float, __clc_fmin, float, float)
+
+#ifdef cl_khr_fp64
+
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+
+_CLC_DEF _CLC_OVERLOAD double __clc_fmin(double x, double y) {
+  x = __builtin_canonicalize(x);
+  y = __builtin_canonicalize(y);
+  return __builtin_fmin(x, y);
+}
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, double, __clc_fmin, double,
+                      double)
+
+#endif
+
+#ifdef cl_khr_fp16
+
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+
+_CLC_DEF _CLC_OVERLOAD half __clc_fmin(half x, half y) {
+  if (__clc_isnan(x))
+    return y;
+  if (__clc_isnan(y))
+    return x;
+  return (y < x) ? y : x;
+}
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, half, __clc_fmin, half, half)
+
+#endif