hypot: Port from amd builtins

v2: Fix whitespace errors

Use only subnormal path.
Passes CTS on carrizo and turks.
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewer: Aaron Watry <awatry@gmail.com>

llvm-svn: 329647

5 files changed, 108 insertions, 4 deletions

diff --git a/libclc/generic/include/math/clc_hypot.h b/libclc/generic/include/math/clc_hypot.h
new file mode 100644
index 000000000000..66901e5d692b
--- /dev/null
+++ b/libclc/generic/include/math/clc_hypot.h
@@ -0,0 +1,5 @@
+#define __CLC_FUNCTION __clc_hypot
+#define __CLC_BODY <clc/math/binary_decl_tt.inc>
+#include <clc/math/gentype.inc>
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
diff --git a/libclc/generic/lib/SOURCES b/libclc/generic/lib/SOURCES
index 222aa1c90ab0..cb2304d00bc3 100644
--- a/libclc/generic/lib/SOURCES
+++ b/libclc/generic/lib/SOURCES
@@ -120,6 +120,7 @@ math/half_rsqrt.cl
 math/half_sin.cl
 math/half_sqrt.cl
 math/half_tan.cl
+math/clc_hypot.cl
 math/hypot.cl
 math/ilogb.cl
 math/clc_ldexp.cl
diff --git a/libclc/generic/lib/math/clc_hypot.cl b/libclc/generic/lib/math/clc_hypot.cl
new file mode 100644
index 000000000000..35532a953206
--- /dev/null
+++ b/libclc/generic/lib/math/clc_hypot.cl
@@ -0,0 +1,98 @@
+/*
+ * Copyright (c) 2014 Advanced Micro Devices, Inc.
+ *
+ * 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.
+ */
+
+#include <clc/clc.h>
+#include <math/clc_hypot.h>
+
+#include "config.h"
+#include "math.h"
+#include "../clcmacro.h"
+
+// Returns sqrt(x*x + y*y) with no overflow or underflow unless the result warrants it
+_CLC_DEF _CLC_OVERLOAD float __clc_hypot(float x, float y)
+{
+    uint ux = as_uint(x);
+    uint aux = ux & EXSIGNBIT_SP32;
+    uint uy = as_uint(y);
+    uint auy = uy & EXSIGNBIT_SP32;
+    float retval;
+    int c = aux > auy;
+    ux = c ? aux : auy;
+    uy = c ? auy : aux;
+
+    int xexp = clamp((int)(ux >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32, -126, 126);
+    float fx_exp = as_float((xexp + EXPBIAS_SP32) << EXPSHIFTBITS_SP32);
+    float fi_exp = as_float((-xexp + EXPBIAS_SP32) << EXPSHIFTBITS_SP32);
+    float fx = as_float(ux) * fi_exp;
+    float fy = as_float(uy) * fi_exp;
+    retval = sqrt(mad(fx, fx, fy*fy)) * fx_exp;
+
+    retval = ux > PINFBITPATT_SP32 | uy == 0 ? as_float(ux) : retval;
+    retval = ux == PINFBITPATT_SP32 | uy == PINFBITPATT_SP32 ? as_float(PINFBITPATT_SP32) : retval;
+    return retval;
+}
+_CLC_BINARY_VECTORIZE(_CLC_DEF _CLC_OVERLOAD, float, __clc_hypot, float, float)
+
+#ifdef cl_khr_fp64
+_CLC_DEF _CLC_OVERLOAD double __clc_hypot(double x, double y)
+{
+    ulong ux = as_ulong(x) & ~SIGNBIT_DP64;
+    int xexp = ux >> EXPSHIFTBITS_DP64;
+    x = as_double(ux);
+
+    ulong uy = as_ulong(y) & ~SIGNBIT_DP64;
+    int yexp = uy >> EXPSHIFTBITS_DP64;
+    y = as_double(uy);
+
+    int c = xexp > EXPBIAS_DP64 + 500 | yexp > EXPBIAS_DP64 + 500;
+    double preadjust = c ? 0x1.0p-600 : 1.0;
+    double postadjust = c ? 0x1.0p+600 : 1.0;
+
+    c = xexp < EXPBIAS_DP64 - 500 | yexp < EXPBIAS_DP64 - 500;
+    preadjust = c ? 0x1.0p+600 : preadjust;
+    postadjust = c ? 0x1.0p-600 : postadjust;
+
+    double ax = x * preadjust;
+    double ay = y * preadjust;
+
+    // The post adjust may overflow, but this can't be avoided in any case
+    double r = sqrt(fma(ax, ax, ay*ay)) * postadjust;
+
+    // If the difference in exponents between x and y is large
+    double s = x + y;
+    c = abs(xexp - yexp) > MANTLENGTH_DP64 + 1;
+    r = c ? s : r;
+
+    // Check for NaN
+    //c = x != x | y != y;
+    c = isnan(x) | isnan(y);
+    r = c ? as_double(QNANBITPATT_DP64) : r;
+
+    // If either is Inf, we must return Inf
+    c = x == as_double(PINFBITPATT_DP64) | y == as_double(PINFBITPATT_DP64);
+    r = c ? as_double(PINFBITPATT_DP64) : r;
+
+    return r;
+}
+
+_CLC_BINARY_VECTORIZE(_CLC_DEF _CLC_OVERLOAD, double, __clc_hypot, double, double)
+#endif
diff --git a/libclc/generic/lib/math/hypot.cl b/libclc/generic/lib/math/hypot.cl
index 8be351e6a981..8339ec75c002 100644
--- a/libclc/generic/lib/math/hypot.cl
+++ b/libclc/generic/lib/math/hypot.cl
@@ -1,4 +1,7 @@
 #include <clc/clc.h>
 
-#define __CLC_BODY <hypot.inc>
+#include <math/clc_hypot.h>
+
+#define __CLC_FUNC hypot
+#define __CLC_BODY <clc_sw_binary.inc>
 #include <clc/math/gentype.inc>
diff --git a/libclc/generic/lib/math/hypot.inc b/libclc/generic/lib/math/hypot.inc
deleted file mode 100644
index 036cee7e1f06..000000000000
--- a/libclc/generic/lib/math/hypot.inc
+++ /dev/null
@@ -1,3 +0,0 @@
-_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE hypot(__CLC_GENTYPE x, __CLC_GENTYPE y) {
-  return sqrt(x*x + y*y);
-}