[libclc] Move fma to the CLC library

This builtin is a little more involved than others as targets deal with
fma in various different ways.

Fundamentally, the CLC __clc_fma builtin compiles to
__builtin_elementwise_fma, which compiles to the @llvm.fma intrinsic.
However, in the case of fp32 fma some targets call the __clc_sw_fma
function, which provides a software implementation of the builtin. This
in principle is controlled by the __CLC_HAVE_HW_FMA32 macro and may be a
runtime decision, depending on how the target defines that macro.

All targets build the CLC fma functions for all types. This is to the
CLC library can have a reliable internal implementation for its own
purposes.

For AMD/NVPTX targets there are no meaningful changes to the generated
LLVM bytecode. Some blocks of code have moved around, which confounds
llvm-diff.

For the clspv and SPIR-V/Mesa targets, only fp32 fma is of interest. Its
use in libclc is tightly controlled by checking __CLC_HAVE_HW_FMA32
first. This can either be a compile-time constant (0, for clspv) or a
runtime function for SPIR-V/Mesa.

The SPIR-V/Mesa target only provided fp32 fma in the OpenCL layer. It
unconditionally mapped that to the __clc_sw_fma builtin, even though the
generic version in theory had a runtime toggle through
__CLC_HAVE_HW_FMA32 specifically for that target. Callers of fma,
though, would end up using the ExtInst fma, *not* calling the _Z3fmafff
function provided by libclc.

This commit keeps this system in place in the OpenCL layer, by mapping
fma to __clc_sw_fma. Where other builtins would previously call fma
(i.e., result in the ExtInst), they now call __clc_fma. This function
checks the __CLC_HAVE_HW_FMA32 runtime toggle, which selects between the
slow version or the quick version. The quick version is the LLVM fma
intrinsic which llvm-spirv translates to the ExtInst.

The clspv target had its own software implementation of fp32 fma, which
it called unconditionally - even though __CLC_HAVE_HW_FMA32 is 1 for
that target. This is potentially just so its library ships a software
version which it can fall back on. In the OpenCL layer, the target
doesn't provide fp64 fma, and maps fp16 fma to fp32 mad.

This commit keeps this system roughly in place: in the OpenCL layer it
maps fp32 fma to __clc_sw_fma, and fp16 fma to mad. Where builtins would
previously call into fma, they now call __clc_fma, which compiles to the
LLVM intrinsic. If this goes through a translation to SPIR-V it will
become the fma ExtInst, or the intrinsic could be replaced by the
_Z3fmafff software implementation.

The clspv and SPIR-V/Mesa targets could potentially be cleaned up later,
depending on their needs.

Author: frasercrmck

libclc/clc/include/clc/float/definitions.h

@@ -0,0 +1,90 @@
+#define MAXFLOAT 0x1.fffffep127f
+#define HUGE_VALF __builtin_huge_valf()
+#define INFINITY __builtin_inff()
+#define NAN __builtin_nanf("")
+
+#define FLT_DIG 6
+#define FLT_MANT_DIG 24
+#define FLT_MAX_10_EXP +38
+#define FLT_MAX_EXP +128
+#define FLT_MIN_10_EXP -37
+#define FLT_MIN_EXP -125
+#define FLT_RADIX 2
+#define FLT_MAX MAXFLOAT
+#define FLT_MIN 0x1.0p-126f
+#define FLT_EPSILON 0x1.0p-23f
+
+#define FP_ILOGB0 (-2147483647 - 1)
+#define FP_ILOGBNAN 2147483647
+
+#define M_E_F 0x1.5bf0a8p+1f
+#define M_LOG2E_F 0x1.715476p+0f
+#define M_LOG10E_F 0x1.bcb7b2p-2f
+#define M_LN2_F 0x1.62e430p-1f
+#define M_LN10_F 0x1.26bb1cp+1f
+#define M_PI_F 0x1.921fb6p+1f
+#define M_PI_2_F 0x1.921fb6p+0f
+#define M_PI_4_F 0x1.921fb6p-1f
+#define M_1_PI_F 0x1.45f306p-2f
+#define M_2_PI_F 0x1.45f306p-1f
+#define M_2_SQRTPI_F 0x1.20dd76p+0f
+#define M_SQRT2_F 0x1.6a09e6p+0f
+#define M_SQRT1_2_F 0x1.6a09e6p-1f
+
+#ifdef __CLC_INTERNAL
+#define M_LOG210_F 0x1.a934f0p+1f
+#endif
+
+#ifdef cl_khr_fp64
+
+#define HUGE_VAL __builtin_huge_val()
+
+#define DBL_DIG 15
+#define DBL_MANT_DIG 53
+#define DBL_MAX_10_EXP +308
+#define DBL_MAX_EXP +1024
+#define DBL_MIN_10_EXP -307
+#define DBL_MIN_EXP -1021
+#define DBL_MAX 0x1.fffffffffffffp1023
+#define DBL_MIN 0x1.0p-1022
+#define DBL_EPSILON 0x1.0p-52
+
+#define M_E 0x1.5bf0a8b145769p+1
+#define M_LOG2E 0x1.71547652b82fep+0
+#define M_LOG10E 0x1.bcb7b1526e50ep-2
+#define M_LN2 0x1.62e42fefa39efp-1
+#define M_LN10 0x1.26bb1bbb55516p+1
+#define M_PI 0x1.921fb54442d18p+1
+#define M_PI_2 0x1.921fb54442d18p+0
+#define M_PI_4 0x1.921fb54442d18p-1
+#define M_1_PI 0x1.45f306dc9c883p-2
+#define M_2_PI 0x1.45f306dc9c883p-1
+#define M_2_SQRTPI 0x1.20dd750429b6dp+0
+#define M_SQRT2 0x1.6a09e667f3bcdp+0
+#define M_SQRT1_2 0x1.6a09e667f3bcdp-1
+
+#ifdef __CLC_INTERNAL
+#define M_LOG210 0x1.a934f0979a371p+1
+#endif
+
+#endif
+
+#ifdef cl_khr_fp16
+
+#if __OPENCL_VERSION__ >= 120
+
+#define HALF_DIG 3
+#define HALF_MANT_DIG 11
+#define HALF_MAX_10_EXP +4
+#define HALF_MAX_EXP +16
+#define HALF_MIN_10_EXP -4
+#define HALF_MIN_EXP -13
+
+#define HALF_RADIX 2
+#define HALF_MAX 0x1.ffcp15h
+#define HALF_MIN 0x1.0p-14h
+#define HALF_EPSILON 0x1.0p-10h
+
+#endif
+
+#endif

libclc/clc/include/clc/integer/clc_abs.h

@@ -1,14 +1,7 @@
 #ifndef __CLC_INTEGER_CLC_ABS_H__
 #define __CLC_INTEGER_CLC_ABS_H__
 
-#if defined(CLC_CLSPV) || defined(CLC_SPIRV)
-// clspv and spir-v targets provide their own OpenCL-compatible abs
-#define __clc_abs abs
-#else
-
 #define __CLC_BODY <clc/integer/clc_abs.inc>
 #include <clc/integer/gentype.inc>
 
-#endif
-
 #endif // __CLC_INTEGER_CLC_ABS_H__

libclc/generic/include/math/clc_fma.h renamed to libclc/clc/include/clc/internal/math/clc_sw_fma.h

@@ -1,11 +1,12 @@
-#define __CLC_FUNCTION __clc_fma
-#define __CLC_INTRINSIC "llvm.fma"
-#include "math/ternary_intrin.inc"
+#ifndef __CLC_INTERNAL_MATH_CLC_SW_FMA_H__
+#define __CLC_INTERNAL_MATH_CLC_SW_FMA_H__
 
-#define __FLOAT_ONLY
 #define __CLC_FUNCTION __clc_sw_fma
 #define __CLC_BODY <clc/shared/ternary_decl.inc>
+
 #include <clc/math/gentype.inc>
+
 #undef __CLC_BODY
 #undef __CLC_FUNCTION
-#undef __FLOAT_ONLY
+
+#endif // __CLC_INTERNAL_MATH_CLC_SW_FMA_H__

libclc/clc/include/clc/math/clc_fma.h

@@ -0,0 +1,12 @@
+#ifndef __CLC_MATH_CLC_FMA_H__
+#define __CLC_MATH_CLC_FMA_H__
+
+#define __CLC_FUNCTION __clc_fma
+#define __CLC_BODY <clc/shared/ternary_decl.inc>
+
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_FMA_H__

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

@@ -39,12 +39,12 @@
 #define PINF 0x200
 
 #if (defined __AMDGCN__ || defined __R600__) && !defined __HAS_FMAF__
-#define HAVE_HW_FMA32() (0)
+#define __CLC_HAVE_HW_FMA32() (0)
 #elif defined(CLC_SPIRV)
 bool __attribute__((noinline)) __clc_runtime_has_hw_fma32(void);
-#define HAVE_HW_FMA32() __clc_runtime_has_hw_fma32()
+#define __CLC_HAVE_HW_FMA32() __clc_runtime_has_hw_fma32()
 #else
-#define HAVE_HW_FMA32() (1)
+#define __CLC_HAVE_HW_FMA32() (1)
 #endif
 
 #define HAVE_BITALIGN() (0)

libclc/clc/include/clc/relational/clc_isinf.h

@@ -1,11 +1,6 @@
 #ifndef __CLC_RELATIONAL_CLC_ISINF_H__
 #define __CLC_RELATIONAL_CLC_ISINF_H__
 
-#if defined(CLC_CLSPV) || defined(CLC_SPIRV)
-// clspv and spir-v targets provide their own OpenCL-compatible isinf
-#define __clc_isinf isinf
-#else
-
 #include <clc/clcfunc.h>
 #include <clc/clctypes.h>
 
@@ -37,6 +32,4 @@ _CLC_VECTOR_ISINF_DECL(short, half)
 #undef _CLC_ISINF_DECL
 #undef _CLC_VECTOR_ISINF_DECL
 
-#endif
-
 #endif // __CLC_RELATIONAL_CLC_ISINF_H__

libclc/clc/include/clc/shared/clc_max.h

@@ -1,17 +1,10 @@
 #ifndef __CLC_SHARED_CLC_MAX_H__
 #define __CLC_SHARED_CLC_MAX_H__
 
-#if defined(CLC_CLSPV) || defined(CLC_SPIRV)
-// clspv and spir-v targets provide their own OpenCL-compatible max
-#define __clc_max max
-#else
-
 #define __CLC_BODY <clc/shared/clc_max.inc>
 #include <clc/integer/gentype.inc>
 
 #define __CLC_BODY <clc/shared/clc_max.inc>
 #include <clc/math/gentype.inc>
 
-#endif
-
 #endif // __CLC_SHARED_CLC_MAX_H__

libclc/clc/include/clc/shared/clc_min.h

@@ -1,17 +1,10 @@
 #ifndef __CLC_SHARED_CLC_MIN_H__
 #define __CLC_SHARED_CLC_MIN_H__
 
-#if defined(CLC_CLSPV) || defined(CLC_SPIRV)
-// clspv and spir-v targets provide their own OpenCL-compatible min
-#define __clc_min min
-#else
-
 #define __CLC_BODY <clc/shared/clc_min.inc>
 #include <clc/integer/gentype.inc>
 
 #define __CLC_BODY <clc/shared/clc_min.inc>
 #include <clc/math/gentype.inc>
 
-#endif
-
 #endif // __CLC_SHARED_CLC_MIN_H__

libclc/clc/lib/clspv/SOURCES

@@ -1,3 +1,4 @@
+../generic/integer/clc_abs.cl
 ../generic/integer/clc_add_sat.cl
 ../generic/integer/clc_clz.cl
 ../generic/integer/clc_hadd.cl
@@ -14,10 +15,15 @@
 ../generic/math/clc_copysign.cl
 ../generic/math/clc_fabs.cl
 ../generic/math/clc_floor.cl
+../generic/math/clc_fma.cl
 ../generic/math/clc_mad.cl
 ../generic/math/clc_nextafter.cl
 ../generic/math/clc_rint.cl
 ../generic/math/clc_trunc.cl
+../generic/relational/clc_isinf.cl
 ../generic/relational/clc_isnan.cl
 ../generic/relational/clc_select.cl
 ../generic/shared/clc_clamp.cl
+../generic/shared/clc_max.cl
+../generic/shared/clc_min.cl
+math/clc_sw_fma.cl