math: Add fdim implementation

Based on the amd-builtin, but explicitly vectorized for all sizes (not just
float4), and includes a vectorized double implementation.

Passes piglit (float) tests on pitcairn.

Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 268708

libclc/generic/include/clc/clc.h

@@ -52,6 +52,7 @@
 #include <clc/math/exp10.h>
 #include <clc/math/exp2.h>
 #include <clc/math/fabs.h>
+#include <clc/math/fdim.h>
 #include <clc/math/floor.h>
 #include <clc/math/fma.h>
 #include <clc/math/fmax.h>

libclc/generic/include/clc/math/fdim.h

@@ -0,0 +1,2 @@
+#define __CLC_BODY <clc/math/fdim.inc>
+#include <clc/math/gentype.inc>

libclc/generic/include/clc/math/fdim.inc

@@ -0,0 +1 @@
+_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE fdim(__CLC_GENTYPE a, __CLC_GENTYPE b);

libclc/generic/lib/SOURCES

@@ -82,6 +82,7 @@ math/exp.cl
 math/exp_helper.cl
 math/exp2.cl
 math/exp10.cl
+math/fdim.cl
 math/fmax.cl
 math/fmin.cl
 math/fmod.cl

libclc/generic/lib/math/fdim.cl

@@ -0,0 +1,10 @@
+#include <clc/clc.h>
+
+#include "math.h"
+
+#ifdef cl_khr_fp64
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+#endif
+
+#define __CLC_BODY <fdim.inc>
+#include <clc/math/gentype.inc>

libclc/generic/lib/math/fdim.inc

@@ -0,0 +1,71 @@
+/*
+ * Copyright (c) 2014 Advanced Micro Devices, Inc.
+ * Copyright (c) 2016 Aaron Watry
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+#if __CLC_FPSIZE == 32
+#ifdef __CLC_SCALAR
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE fdim(__CLC_GENTYPE x, __CLC_GENTYPE y) {
+    if (__builtin_isnan(x) || __builtin_isnan(y))
+        return as_float(QNANBITPATT_SP32);
+    return __builtin_fmax(x - y, 0);
+}
+#define __CLC_FDIM_VEC(width) \
+_CLC_OVERLOAD _CLC_DEF float##width fdim(float##width x, float##width y) { \
+    /* Determine if x or y is NaN. */ \
+    /* Vector true is -1, i.e. all-bits-set, and NaN==NaN is false. */ \
+    /* If either is NaN, then ~((x==x) & (y==y)) will be 0 (e.g. ~(-1)), as will n. */ \
+    int##width n = ~((x == x) & (y == y)) & QNANBITPATT_SP32; \
+    /* Calculate x-y if x>y, otherwise positive 0, again taking */ \
+    /* advantage of vector true being all-bits-set. */ \
+    int##width r = (x > y) & as_int##width(x - y); \
+    return as_float##width(n | r); \
+}
+__CLC_FDIM_VEC(2)
+__CLC_FDIM_VEC(3)
+__CLC_FDIM_VEC(4)
+__CLC_FDIM_VEC(8)
+__CLC_FDIM_VEC(16)
+#undef __CLC_FDIM_VEC
+#endif
+#endif
+
+#if __CLC_FPSIZE == 64
+#ifdef __CLC_SCALAR
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE fdim(__CLC_GENTYPE x, private __CLC_GENTYPE y) {
+    long n = -(isnan(x) | isnan(y)) & QNANBITPATT_DP64;
+    long r = -(x > y) & as_long(x - y);
+    return as_double(n | r);
+}
+#define __CLC_FDIM_VEC(width) \
+_CLC_OVERLOAD _CLC_DEF double##width fdim(double##width x, double##width y) { \
+    /* See comment in float implementation for explanation. */ \
+    long##width n = ~((x == x) & (y == y)) & QNANBITPATT_DP64; \
+    long##width r = (x > y) & as_long##width(x - y); \
+    return as_double##width(n | r); \
+}
+__CLC_FDIM_VEC(2)
+__CLC_FDIM_VEC(3)
+__CLC_FDIM_VEC(4)
+__CLC_FDIM_VEC(8)
+__CLC_FDIM_VEC(16)
+#undef __CLC_FDIM_VEC
+#endif
+#endif