Rebaseusers/pcc/spr/codegen-introduce-machinefunctiongetpreferredalignment

Created using spr 1.3.6-beta.1

23 files changed, 143 insertions, 334 deletions

diff --git a/libclc/CMakeLists.txt b/libclc/CMakeLists.txt
index d060b71..ec4c253 100644
--- a/libclc/CMakeLists.txt
+++ b/libclc/CMakeLists.txt
@@ -110,12 +110,10 @@ else()
   # in-tree build we place the libraries in clang's resource driectory.
   include(GetClangResourceDir)
   get_clang_resource_dir( LIBCLC_INSTALL_DIR )
-  cmake_path( APPEND LIBCLC_INSTALL_DIR "lib" "libclc" )
+  cmake_path( APPEND LIBCLC_INSTALL_DIR "lib" )
 
-  # Note we do not adhere to LLVM_ENABLE_PER_TARGET_RUNTIME_DIR.
   cmake_path( GET LLVM_LIBRARY_OUTPUT_INTDIR PARENT_PATH LIBCLC_OUTPUT_LIBRARY_DIR )
   cmake_path( APPEND LIBCLC_OUTPUT_LIBRARY_DIR ${LIBCLC_INSTALL_DIR} )
-  file( MAKE_DIRECTORY ${LIBCLC_OUTPUT_LIBRARY_DIR} )
 endif()
 
 if( EXISTS ${LIBCLC_CUSTOM_LLVM_TOOLS_BINARY_DIR} )
@@ -232,6 +230,7 @@ set( tahiti_aliases pitcairn verde oland hainan bonaire kabini kaveri hawaii
   gfx1100 gfx1101 gfx1102 gfx1103
   gfx1150 gfx1151 gfx1152 gfx1153
   gfx1200 gfx1201 gfx1250 gfx1251
+  gfx1310
 )
 
 # pkg-config file
@@ -408,6 +407,7 @@ foreach( t ${LIBCLC_TARGETS_TO_BUILD} )
       -D${CLC_TARGET_DEFINE}
       # All libclc builtin libraries see CLC headers
       -I${CMAKE_CURRENT_SOURCE_DIR}/clc/include
+      -include opencl-c-base.h
       # Error on undefined macros
       -Werror=undef
       -fdiscard-value-names
@@ -445,6 +445,7 @@ foreach( t ${LIBCLC_TARGETS_TO_BUILD} )
     add_libclc_builtin_set(
       CLC_INTERNAL
       ARCH ${ARCH}
+      DEVICE ${d}
       ARCH_SUFFIX clc-${arch_suffix}
       TRIPLE ${clang_triple}
       COMPILE_FLAGS ${build_flags}
@@ -453,18 +454,20 @@ foreach( t ${LIBCLC_TARGETS_TO_BUILD} )
     )
 
     list( APPEND build_flags
-      -Xclang -fdeclare-opencl-builtins -Xclang -finclude-default-header
+      -Xclang -fdeclare-opencl-builtins
       -I${CMAKE_CURRENT_SOURCE_DIR}/opencl/include
     )
 
     add_libclc_builtin_set(
       ARCH ${ARCH}
+      DEVICE ${d}
       ARCH_SUFFIX ${arch_suffix}
       TRIPLE ${clang_triple}
       COMPILE_FLAGS ${build_flags}
       OPT_FLAGS ${opt_flags}
       LIB_FILES ${opencl_lib_files}
       ALIASES ${${d}_aliases}
+      OUTPUT_FILENAME libclc
       PARENT_TARGET libclc-opencl-builtins
       # Link in the CLC builtins and internalize their symbols
       INTERNAL_LINK_DEPENDENCIES builtins.link.clc-${arch_suffix}
diff --git a/libclc/clc/include/clc/clctypes.h b/libclc/clc/include/clc/clctypes.h
deleted file mode 100644
index 2187c25..0000000
--- a/libclc/clc/include/clc/clctypes.h
+++ /dev/null
@@ -1,108 +0,0 @@
-//===----------------------------------------------------------------------===//
-//
-// 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_CLCTYPES_H_
-#define __CLC_CLCTYPES_H_
-
-/* 6.1.1 Built-in Scalar Data Types */
-
-typedef unsigned char uchar;
-typedef unsigned short ushort;
-typedef unsigned int uint;
-typedef unsigned long ulong;
-
-typedef __SIZE_TYPE__ size_t;
-typedef __PTRDIFF_TYPE__ ptrdiff_t;
-
-#define __stdint_join3(a, b, c) a##b##c
-
-#define __intn_t(n) __stdint_join3(__INT, n, _TYPE__)
-#define __uintn_t(n) __stdint_join3(unsigned __INT, n, _TYPE__)
-
-typedef __intn_t(__INTPTR_WIDTH__) intptr_t;
-typedef __uintn_t(__INTPTR_WIDTH__) uintptr_t;
-
-#undef __uintn_t
-#undef __intn_t
-#undef __stdint_join3
-
-/* 6.1.2 Built-in Vector Data Types */
-
-typedef __attribute__((ext_vector_type(2))) char char2;
-typedef __attribute__((ext_vector_type(3))) char char3;
-typedef __attribute__((ext_vector_type(4))) char char4;
-typedef __attribute__((ext_vector_type(8))) char char8;
-typedef __attribute__((ext_vector_type(16))) char char16;
-
-typedef __attribute__((ext_vector_type(2))) uchar uchar2;
-typedef __attribute__((ext_vector_type(3))) uchar uchar3;
-typedef __attribute__((ext_vector_type(4))) uchar uchar4;
-typedef __attribute__((ext_vector_type(8))) uchar uchar8;
-typedef __attribute__((ext_vector_type(16))) uchar uchar16;
-
-typedef __attribute__((ext_vector_type(2))) short short2;
-typedef __attribute__((ext_vector_type(3))) short short3;
-typedef __attribute__((ext_vector_type(4))) short short4;
-typedef __attribute__((ext_vector_type(8))) short short8;
-typedef __attribute__((ext_vector_type(16))) short short16;
-
-typedef __attribute__((ext_vector_type(2))) ushort ushort2;
-typedef __attribute__((ext_vector_type(3))) ushort ushort3;
-typedef __attribute__((ext_vector_type(4))) ushort ushort4;
-typedef __attribute__((ext_vector_type(8))) ushort ushort8;
-typedef __attribute__((ext_vector_type(16))) ushort ushort16;
-
-typedef __attribute__((ext_vector_type(2))) int int2;
-typedef __attribute__((ext_vector_type(3))) int int3;
-typedef __attribute__((ext_vector_type(4))) int int4;
-typedef __attribute__((ext_vector_type(8))) int int8;
-typedef __attribute__((ext_vector_type(16))) int int16;
-
-typedef __attribute__((ext_vector_type(2))) uint uint2;
-typedef __attribute__((ext_vector_type(3))) uint uint3;
-typedef __attribute__((ext_vector_type(4))) uint uint4;
-typedef __attribute__((ext_vector_type(8))) uint uint8;
-typedef __attribute__((ext_vector_type(16))) uint uint16;
-
-typedef __attribute__((ext_vector_type(2))) long long2;
-typedef __attribute__((ext_vector_type(3))) long long3;
-typedef __attribute__((ext_vector_type(4))) long long4;
-typedef __attribute__((ext_vector_type(8))) long long8;
-typedef __attribute__((ext_vector_type(16))) long long16;
-
-typedef __attribute__((ext_vector_type(2))) ulong ulong2;
-typedef __attribute__((ext_vector_type(3))) ulong ulong3;
-typedef __attribute__((ext_vector_type(4))) ulong ulong4;
-typedef __attribute__((ext_vector_type(8))) ulong ulong8;
-typedef __attribute__((ext_vector_type(16))) ulong ulong16;
-
-typedef __attribute__((ext_vector_type(2))) float float2;
-typedef __attribute__((ext_vector_type(3))) float float3;
-typedef __attribute__((ext_vector_type(4))) float float4;
-typedef __attribute__((ext_vector_type(8))) float float8;
-typedef __attribute__((ext_vector_type(16))) float float16;
-
-/* 9.3 Double Precision Floating-Point */
-
-#ifdef cl_khr_fp64
-typedef __attribute__((ext_vector_type(2))) double double2;
-typedef __attribute__((ext_vector_type(3))) double double3;
-typedef __attribute__((ext_vector_type(4))) double double4;
-typedef __attribute__((ext_vector_type(8))) double double8;
-typedef __attribute__((ext_vector_type(16))) double double16;
-#endif
-
-#ifdef cl_khr_fp16
-typedef __attribute__((ext_vector_type(2))) half half2;
-typedef __attribute__((ext_vector_type(3))) half half3;
-typedef __attribute__((ext_vector_type(4))) half half4;
-typedef __attribute__((ext_vector_type(8))) half half8;
-typedef __attribute__((ext_vector_type(16))) half half16;
-#endif
-
-#endif // __CLC_CLCTYPES_H_
diff --git a/libclc/clc/include/clc/float/definitions.h b/libclc/clc/include/clc/float/definitions.h
index 0317bf1..25f35a4 100644
--- a/libclc/clc/include/clc/float/definitions.h
+++ b/libclc/clc/include/clc/float/definitions.h
@@ -6,16 +6,6 @@
 //
 //===----------------------------------------------------------------------===//
 
-#ifndef MAXFLOAT
-#define MAXFLOAT 0x1.fffffep127f
-#endif
-#ifndef HUGE_VALF
-#define HUGE_VALF __builtin_huge_valf()
-#endif
-#ifndef INFINITY
-#define INFINITY __builtin_inff()
-#endif
-
 #define FLT_DIG 6
 #define FLT_MANT_DIG 24
 #define FLT_MAX_10_EXP +38
@@ -23,68 +13,14 @@
 #define FLT_MIN_10_EXP -37
 #define FLT_MIN_EXP -125
 #define FLT_RADIX 2
-#ifndef FLT_MAX
-#define FLT_MAX MAXFLOAT
-#endif
 #define FLT_MIN 0x1.0p-126f
 #define FLT_EPSILON 0x1.0p-23f
 #define FLT_NAN __builtin_nanf("")
 
-#ifndef FP_ILOGB0
-#define FP_ILOGB0 (-2147483647 - 1)
-#endif
-#ifndef FP_ILOGBNAN
-#define FP_ILOGBNAN 2147483647
-#endif
-
-#ifndef M_E_F
-#define M_E_F 0x1.5bf0a8p+1f
-#endif
-#ifndef M_LOG2E_F
-#define M_LOG2E_F 0x1.715476p+0f
-#endif
-#ifndef M_LOG10E_F
-#define M_LOG10E_F 0x1.bcb7b2p-2f
-#endif
-#ifndef M_LN2_F
-#define M_LN2_F 0x1.62e430p-1f
-#endif
-#ifndef M_LN10_F
-#define M_LN10_F 0x1.26bb1cp+1f
-#endif
-#ifndef M_PI_F
-#define M_PI_F 0x1.921fb6p+1f
-#endif
-#ifndef M_PI_2_F
-#define M_PI_2_F 0x1.921fb6p+0f
-#endif
-#ifndef M_PI_4_F
-#define M_PI_4_F 0x1.921fb6p-1f
-#endif
-#ifndef M_1_PI_F
-#define M_1_PI_F 0x1.45f306p-2f
-#endif
-#ifndef M_2_PI_F
-#define M_2_PI_F 0x1.45f306p-1f
-#endif
-#ifndef M_2_SQRTPI_F
-#define M_2_SQRTPI_F 0x1.20dd76p+0f
-#endif
-#ifndef M_SQRT2_F
-#define M_SQRT2_F 0x1.6a09e6p+0f
-#endif
-#ifndef M_SQRT1_2_F
-#define M_SQRT1_2_F 0x1.6a09e6p-1f
-#endif
-
 #define M_LOG210_F 0x1.a934f0p+1f
 
 #ifdef cl_khr_fp64
 
-#ifndef HUGE_VAL
-#define HUGE_VAL __builtin_huge_val()
-#endif
-
 #define DBL_DIG 15
 #define DBL_MANT_DIG 53
 #define DBL_MAX_10_EXP +308
@@ -120,21 +56,7 @@
 #define HALF_MAX_EXP +16
 #define HALF_MIN_10_EXP -4
 #define HALF_MIN_EXP -13
-
 #define HALF_RADIX 2
-#ifndef HALF_MAX
-#define HALF_MAX 0x1.ffcp15h
-#endif
-#ifndef HALF_MIN
-#define HALF_MIN 0x1.0p-14h
-#endif
-#ifndef HALF_EPSILON
-#define HALF_EPSILON 0x1.0p-10h
-#endif
 #define HALF_NAN __builtin_nanf16("")
 
-#ifndef M_LOG2E_H
-#define M_LOG2E_H 0x1.714p+0h
-#endif
-
 #endif
diff --git a/libclc/clc/include/clc/integer/clc_upsample.h b/libclc/clc/include/clc/integer/clc_upsample.h
index 51728df..ca3b8a9 100644
--- a/libclc/clc/include/clc/integer/clc_upsample.h
+++ b/libclc/clc/include/clc/integer/clc_upsample.h
@@ -10,7 +10,6 @@
 #define __CLC_INTEGER_CLC_UPSAMPLE_H__
 
 #include <clc/clcfunc.h>
-#include <clc/clctypes.h>
 
 #define __CLC_UPSAMPLE_DECL(BGENTYPE, GENTYPE, UGENTYPE)                       \
   _CLC_OVERLOAD _CLC_DECL BGENTYPE __clc_upsample(GENTYPE hi, UGENTYPE lo);
diff --git a/libclc/clc/include/clc/integer/definitions.h b/libclc/clc/include/clc/integer/definitions.h
index 3b9d57f..3ef8ad2 100644
--- a/libclc/clc/include/clc/integer/definitions.h
+++ b/libclc/clc/include/clc/integer/definitions.h
@@ -11,23 +11,11 @@
 
 #define CHAR_BIT 8
 #define INT_MAX 2147483647
-#ifndef INT_MIN
-#define INT_MIN (-2147483647 - 1)
-#endif
 #define LONG_MAX 0x7fffffffffffffffL
-#ifndef LONG_MIN
-#define LONG_MIN (-0x7fffffffffffffffL - 1)
-#endif
 #define CHAR_MAX SCHAR_MAX
 #define CHAR_MIN SCHAR_MIN
 #define SCHAR_MAX 127
-#ifndef SCHAR_MIN
-#define SCHAR_MIN (-127 - 1)
-#endif
 #define SHRT_MAX 32767
-#ifndef SHRT_MIN
-#define SHRT_MIN (-32767 - 1)
-#endif
 #define UCHAR_MAX 255
 #define UCHAR_MIN 0
 #define USHRT_MAX 65535
diff --git a/libclc/clc/include/clc/integer/gentype.inc b/libclc/clc/include/clc/integer/gentype.inc
index acca863..36d7ae9 100644
--- a/libclc/clc/include/clc/integer/gentype.inc
+++ b/libclc/clc/include/clc/integer/gentype.inc
@@ -7,7 +7,6 @@
 //===----------------------------------------------------------------------===//
 
 #include <clc/clcfunc.h>
-#include <clc/clctypes.h>
 #include <clc/utils.h>
 
 #define __CLC_AS_GENTYPE __CLC_XCONCAT(__clc_as_, __CLC_GENTYPE)
diff --git a/libclc/clc/include/clc/integer/gentype24.inc b/libclc/clc/include/clc/integer/gentype24.inc
index b787f25..c74ce0c 100644
--- a/libclc/clc/include/clc/integer/gentype24.inc
+++ b/libclc/clc/include/clc/integer/gentype24.inc
@@ -7,7 +7,6 @@
 //===----------------------------------------------------------------------===//
 
 #include <clc/clcfunc.h>
-#include <clc/clctypes.h>
 
 #define __CLC_GENSIZE 32
 #undef __CLC_SCALAR_GENTYPE
diff --git a/libclc/clc/include/clc/internal/clc.h b/libclc/clc/include/clc/internal/clc.h
index d9fb481..fcfb223 100644
--- a/libclc/clc/include/clc/internal/clc.h
+++ b/libclc/clc/include/clc/internal/clc.h
@@ -27,7 +27,6 @@
 #include <clc/clcfunc.h>
 
 /* 6.1 Supported Data Types */
-#include <clc/clctypes.h>
 
 /* 6.2.4.2 Reinterpreting Types Using __clc_as_type() and __clc_as_typen() */
 #include <clc/clc_as_type.h>
diff --git a/libclc/clc/include/clc/math/gentype.inc b/libclc/clc/include/clc/math/gentype.inc
index 3373f5f..59626d6 100644
--- a/libclc/clc/include/clc/math/gentype.inc
+++ b/libclc/clc/include/clc/math/gentype.inc
@@ -7,7 +7,6 @@
 //===----------------------------------------------------------------------===//
 
 #include <clc/clcfunc.h>
-#include <clc/clctypes.h>
 #include <clc/utils.h>
 
 // Define some useful macros for type conversions.
diff --git a/libclc/clc/include/clc/math/tables.h b/libclc/clc/include/clc/math/tables.h
index ad4b054..12361a3 100644
--- a/libclc/clc/include/clc/math/tables.h
+++ b/libclc/clc/include/clc/math/tables.h
@@ -9,8 +9,6 @@
 #ifndef __CLC_MATH_TABLES_H__
 #define __CLC_MATH_TABLES_H__
 
-#include <clc/clctypes.h>
-
 #define __CLC_TABLE_SPACE __constant
 
 #define __CLC_TABLE_MANGLE(NAME) __clc_##NAME
diff --git a/libclc/clc/include/clc/relational/clc_all.h b/libclc/clc/include/clc/relational/clc_all.h
index 272d879..a99e2ae 100644
--- a/libclc/clc/include/clc/relational/clc_all.h
+++ b/libclc/clc/include/clc/relational/clc_all.h
@@ -10,7 +10,6 @@
 #define __CLC_RELATIONAL_CLC_ALL_H__
 
 #include <clc/clcfunc.h>
-#include <clc/clctypes.h>
 
 #define _CLC_ALL_DECL(TYPE)                                                    \
   _CLC_OVERLOAD _CLC_CONST _CLC_DECL int __clc_all(TYPE v);
diff --git a/libclc/clc/include/clc/relational/clc_any.h b/libclc/clc/include/clc/relational/clc_any.h
index 82d08c0..5e52be4 100644
--- a/libclc/clc/include/clc/relational/clc_any.h
+++ b/libclc/clc/include/clc/relational/clc_any.h
@@ -10,7 +10,6 @@
 #define __CLC_RELATIONAL_CLC_ANY_H__
 
 #include <clc/clcfunc.h>
-#include <clc/clctypes.h>
 
 #define _CLC_ANY_DECL(TYPE)                                                    \
   _CLC_OVERLOAD _CLC_CONST _CLC_DECL int __clc_any(TYPE v);
diff --git a/libclc/clc/include/clc/relational/clc_isequal.h b/libclc/clc/include/clc/relational/clc_isequal.h
index 2585dbd..18e0538 100644
--- a/libclc/clc/include/clc/relational/clc_isequal.h
+++ b/libclc/clc/include/clc/relational/clc_isequal.h
@@ -10,7 +10,6 @@
 #define __CLC_RELATIONAL_CLC_ISEQUAL_H__
 
 #include <clc/clcfunc.h>
-#include <clc/clctypes.h>
 
 #define _CLC_ISEQUAL_DECL(TYPE, RETTYPE)                                       \
   _CLC_OVERLOAD _CLC_CONST _CLC_DECL RETTYPE __clc_isequal(TYPE x, TYPE y);
diff --git a/libclc/clc/include/clc/relational/clc_isinf.h b/libclc/clc/include/clc/relational/clc_isinf.h
index fcd87e94..a64b18bd 100644
--- a/libclc/clc/include/clc/relational/clc_isinf.h
+++ b/libclc/clc/include/clc/relational/clc_isinf.h
@@ -10,7 +10,6 @@
 #define __CLC_RELATIONAL_CLC_ISINF_H__
 
 #include <clc/clcfunc.h>
-#include <clc/clctypes.h>
 
 #define _CLC_ISINF_DECL(RET_TYPE, ARG_TYPE)                                    \
   _CLC_OVERLOAD _CLC_CONST _CLC_DECL RET_TYPE __clc_isinf(ARG_TYPE);
diff --git a/libclc/clc/include/clc/relational/clc_isnan.h b/libclc/clc/include/clc/relational/clc_isnan.h
index 779e6cc..48cfa7f 100644
--- a/libclc/clc/include/clc/relational/clc_isnan.h
+++ b/libclc/clc/include/clc/relational/clc_isnan.h
@@ -10,7 +10,6 @@
 #define __CLC_RELATIONAL_CLC_ISNAN_H__
 
 #include <clc/clcfunc.h>
-#include <clc/clctypes.h>
 
 #define _CLC_ISNAN_DECL(RET_TYPE, ARG_TYPE)                                    \
   _CLC_OVERLOAD _CLC_CONST _CLC_DECL RET_TYPE __clc_isnan(ARG_TYPE);
diff --git a/libclc/clc/lib/generic/atomic/clc_atomic_dec.cl b/libclc/clc/lib/generic/atomic/clc_atomic_dec.cl
index e0d2419..4603e7e 100644
--- a/libclc/clc/lib/generic/atomic/clc_atomic_dec.cl
+++ b/libclc/clc/lib/generic/atomic/clc_atomic_dec.cl
@@ -9,7 +9,7 @@
 #include <clc/atomic/clc_atomic_dec.h>
 
 #define __CLC_FUNCTION __clc_atomic_dec
-#define __CLC_IMPL_FUNCTION __scoped_atomic_udec_wrap
+#define __CLC_IMPL_FUNCTION __scoped_atomic_fetch_udec
 #define __CLC_INC_DEC
 
 #define __CLC_BODY <clc_atomic_def.inc>
diff --git a/libclc/clc/lib/generic/atomic/clc_atomic_inc.cl b/libclc/clc/lib/generic/atomic/clc_atomic_inc.cl
index d594754..53915c5 100644
--- a/libclc/clc/lib/generic/atomic/clc_atomic_inc.cl
+++ b/libclc/clc/lib/generic/atomic/clc_atomic_inc.cl
@@ -9,7 +9,7 @@
 #include <clc/atomic/clc_atomic_inc.h>
 
 #define __CLC_FUNCTION __clc_atomic_inc
-#define __CLC_IMPL_FUNCTION __scoped_atomic_uinc_wrap
+#define __CLC_IMPL_FUNCTION __scoped_atomic_fetch_uinc
 #define __CLC_INC_DEC
 
 #define __CLC_BODY <clc_atomic_def.inc>
diff --git a/libclc/clc/lib/generic/math/clc_remquo.cl b/libclc/clc/lib/generic/math/clc_remquo.cl
index fd83ead..db7ab8c 100644
--- a/libclc/clc/lib/generic/math/clc_remquo.cl
+++ b/libclc/clc/lib/generic/math/clc_remquo.cl
@@ -7,14 +7,23 @@
 //===----------------------------------------------------------------------===//
 
 #include <clc/clc_convert.h>
+#include <clc/float/definitions.h>
 #include <clc/integer/clc_clz.h>
 #include <clc/internal/clc.h>
+#include <clc/math/clc_copysign.h>
+#include <clc/math/clc_fabs.h>
 #include <clc/math/clc_floor.h>
 #include <clc/math/clc_fma.h>
+#include <clc/math/clc_frexp.h>
 #include <clc/math/clc_ldexp.h>
-#include <clc/math/clc_subnormal_config.h>
+#include <clc/math/clc_nan.h>
+#include <clc/math/clc_native_recip.h>
+#include <clc/math/clc_rint.h>
+#include <clc/math/clc_sincos_helpers.h>
 #include <clc/math/clc_trunc.h>
 #include <clc/math/math.h>
+#include <clc/relational/clc_isfinite.h>
+#include <clc/relational/clc_isnan.h>
 #include <clc/shared/clc_max.h>
 
 #define __CLC_ADDRESS_SPACE private
diff --git a/libclc/clc/lib/generic/math/clc_remquo.inc b/libclc/clc/lib/generic/math/clc_remquo.inc
index 3a76ffe..79eef07 100644
--- a/libclc/clc/lib/generic/math/clc_remquo.inc
+++ b/libclc/clc/lib/generic/math/clc_remquo.inc
@@ -8,69 +8,82 @@
 
 _CLC_DEF _CLC_OVERLOAD float __clc_remquo(float x, float y,
                                           __CLC_ADDRESS_SPACE int *quo) {
-  x = __clc_flush_denormal_if_not_supported(x);
-  y = __clc_flush_denormal_if_not_supported(y);
-  int ux = __clc_as_int(x);
-  int ax = ux & EXSIGNBIT_SP32;
-  float xa = __clc_as_float(ax);
-  int sx = ux ^ ax;
-  int ex = ax >> EXPSHIFTBITS_SP32;
-
-  int uy = __clc_as_int(y);
-  int ay = uy & EXSIGNBIT_SP32;
-  float ya = __clc_as_float(ay);
-  int sy = uy ^ ay;
-  int ey = ay >> EXPSHIFTBITS_SP32;
-
-  float xr = __clc_as_float(0x3f800000 | (ax & 0x007fffff));
-  float yr = __clc_as_float(0x3f800000 | (ay & 0x007fffff));
-  int c;
-  int k = ex - ey;
-
-  uint q = 0;
-
-  while (k > 0) {
-    c = xr >= yr;
-    q = (q << 1) | c;
-    xr -= c ? yr : 0.0f;
-    xr += xr;
-    --k;
+  const int bits = 12;
+  float ax = __clc_fabs(x);
+  float ay = __clc_fabs(y);
+  float ret;
+  int q7;
+  if (ax > ay) {
+    int ex, ey;
+    {
+      int _exp;
+      float _mant = __clc_frexp(ax, &_exp);
+      ex = _exp - 1;
+      ax = __clc_ldexp(_mant, bits);
+    }
+    {
+      int _exp;
+      float _mant = __clc_frexp(ay, &_exp);
+      ey = _exp - 1;
+      ay = __clc_ldexp(_mant, 1);
+    }
+    int nb = ex - ey;
+    float ayinv = __clc_native_recip(ay);
+    int qacc = 0;
+    while (nb > bits) {
+      float q = __clc_rint(ax * ayinv);
+      ax = __clc_fma(-q, ay, ax);
+      int clt = ax < 0.0f;
+      float axp = ax + ay;
+      ax = clt ? axp : ax;
+
+      int iq = (int)q;
+      iq -= clt;
+      qacc = (qacc << bits) | iq;
+
+      ax = __clc_ldexp(ax, bits);
+      nb -= bits;
+    }
+    ax = __clc_ldexp(ax, nb - bits + 1);
+    {
+      float q = __clc_rint(ax * ayinv);
+      ax = __clc_fma(-q, ay, ax);
+      int clt = ax < 0.0f;
+      float axp = ax + ay;
+      ax = clt ? axp : ax;
+      int iq = (int)q;
+      iq -= clt;
+      qacc = (qacc << (nb + 1)) | iq;
+    }
+    int aq = (2.0f * ax > ay) | ((qacc & 0x1) & (2.0f * ax == ay));
+    ax = ax - (aq ? ay : 0.0f);
+    qacc += aq;
+    int qneg = (__clc_as_int(x) ^ __clc_as_int(y)) >> 31;
+    q7 = ((qacc & 0x7f) ^ qneg) - qneg;
+    ax = __clc_ldexp(ax, ey);
+    ret =
+        __clc_as_float((__clc_as_int(x) & (int)0x80000000) ^ __clc_as_int(ax));
+  } else {
+    ret = x;
+    q7 = 0;
+    bool c = (ay < 0x1.0p+127f & 2.0f * ax > ay) | (ax > 0.5f * ay);
+
+    int qsgn = 1 + (((__clc_as_int(x) ^ __clc_as_int(y)) >> 31) << 1);
+    float t = __clc_fma(y, -(float)qsgn, x);
+    ret = c ? t : __builtin_elementwise_canonicalize(x);
+    q7 = c ? qsgn : q7;
+    ret = ax == ay ? __clc_copysign(0.0f, x) : ret;
+    q7 = ax == ay ? qsgn : q7;
   }
 
-  c = xr > yr;
-  q = (q << 1) | c;
-  xr -= c ? yr : 0.0f;
+  ret = y == 0.0f ? __clc_nan(0) : ret;
+  q7 = y == 0.0f ? 0 : q7;
+  bool c = !__clc_isnan(y) && __clc_isfinite(x);
+  ret = c ? ret : __clc_nan(0);
+  q7 = c ? q7 : 0;
 
-  int lt = ex < ey;
-
-  q = lt ? 0 : q;
-  xr = lt ? xa : xr;
-  yr = lt ? ya : yr;
-
-  c = (yr < 2.0f * xr) | ((yr == 2.0f * xr) & ((q & 0x1) == 0x1));
-  xr -= c ? yr : 0.0f;
-  q += c;
-
-  float s = __clc_as_float(ey << EXPSHIFTBITS_SP32);
-  xr *= lt ? 1.0f : s;
-
-  int qsgn = sx == sy ? 1 : -1;
-  int quot = (q & 0x7f) * qsgn;
-
-  c = ax == ay;
-  quot = c ? qsgn : quot;
-  xr = c ? 0.0f : xr;
-
-  xr = __clc_as_float(sx ^ __clc_as_int(xr));
-
-  c = ax > PINFBITPATT_SP32 | ay > PINFBITPATT_SP32 | ax == PINFBITPATT_SP32 |
-      ay == 0;
-  quot = c ? 0 : quot;
-  xr = c ? __clc_as_float(QNANBITPATT_SP32) : xr;
-
-  *quo = quot;
-
-  return xr;
+  *quo = q7;
+  return ret;
 }
 
 // remquo signature is special, we don't have macro for this
diff --git a/libclc/cmake/modules/AddLibclc.cmake b/libclc/cmake/modules/AddLibclc.cmake
index 1142963..7092457b 100644
--- a/libclc/cmake/modules/AddLibclc.cmake
+++ b/libclc/cmake/modules/AddLibclc.cmake
@@ -33,9 +33,7 @@ function(compile_to_bc)
   else()
     set( PP_OPTS -E;-P )
     set( TMP_SUFFIX .tmp )
-    string( REPLACE "-Xclang;-fdeclare-opencl-builtins;-Xclang;-finclude-default-header"
-      "" EXTRA_OPTS "${ARG_EXTRA_OPTS}"
-    )
+    string( REPLACE "-include;opencl-c-base.h" "" EXTRA_OPTS "${ARG_EXTRA_OPTS}")
   endif()
 
   set( TARGET_ARG )
@@ -226,33 +224,6 @@ function(get_libclc_device_info)
   endif()
 endfunction()
 
-# Install libclc artifacts.
-#
-# Arguments:
-#  * FILES <string> ...
-#      List of libclc artifact files to be installed.
-function(libclc_install)
-  cmake_parse_arguments(ARG "" "" "FILES" ${ARGN})
-
-  if( NOT ARG_FILES )
-    message( FATAL_ERROR "Must provide FILES" )
-  endif()
-
-  if( NOT CMAKE_CFG_INTDIR STREQUAL "." )
-    # Replace CMAKE_CFG_INTDIR with CMAKE_INSTALL_CONFIG_NAME for multiple-
-    # configuration generators.
-    string( REPLACE ${CMAKE_CFG_INTDIR} "\$\{CMAKE_INSTALL_CONFIG_NAME\}"
-            files ${ARG_FILES} )
-  else()
-    set( files ${ARG_FILES} )
-  endif()
-
-  install(
-    FILES ${files}
-    DESTINATION ${LIBCLC_INSTALL_DIR}
-  )
-endfunction()
-
 # Compiles a list of library source files (provided by LIB_FILES) and compiles
 # them to LLVM bytecode (or SPIR-V), links them together and optimizes them.
 #
@@ -262,10 +233,14 @@ endfunction()
 # Arguments:
 #  * ARCH <string>
 #      libclc architecture being built
+#  * DEVICE <string>
+#      libclc microarchitecture being built
 #  * ARCH_SUFFIX <string>
 #      libclc architecture/triple suffix
 #  * TRIPLE <string>
 #      Triple used to compile
+#  * OUTPUT_FILENAME <string>
+#      libclc output library name
 #  * PARENT_TARGET <string>
 #      Target into which to group the target builtins
 #
@@ -288,7 +263,7 @@ endfunction()
 function(add_libclc_builtin_set)
   cmake_parse_arguments(ARG
     "CLC_INTERNAL"
-    "ARCH;TRIPLE;ARCH_SUFFIX;PARENT_TARGET"
+    "ARCH;DEVICE;TRIPLE;ARCH_SUFFIX;OUTPUT_FILENAME;PARENT_TARGET"
     "LIB_FILES;COMPILE_FLAGS;OPT_FLAGS;ALIASES;INTERNAL_LINK_DEPENDENCIES"
     ${ARGN}
   )
@@ -399,16 +374,27 @@ function(add_libclc_builtin_set)
     return()
   endif()
 
+  if (NOT DEFINED ARG_OUTPUT_FILENAME OR ARG_OUTPUT_FILENAME STREQUAL "")
+    message(FATAL_ERROR "OUTPUT_FILENAME parameter is required and must be non-empty.")
+  endif()
+  set( LIBCLC_OUTPUT_FILENAME ${ARG_OUTPUT_FILENAME} )
   set( builtins_link_lib $<TARGET_PROPERTY:${builtins_link_lib_tgt},TARGET_FILE> )
 
+  # We store the library according to its triple and cpu if present.
+  if (NOT "${ARG_DEVICE}" STREQUAL "none")
+    set (library_dir ${LIBCLC_OUTPUT_LIBRARY_DIR}/${ARG_TRIPLE}/${ARG_DEVICE})
+  else()
+    set (library_dir ${LIBCLC_OUTPUT_LIBRARY_DIR}/${ARG_TRIPLE})
+  endif()
+  file( MAKE_DIRECTORY ${library_dir} )
+
   # For SPIR-V targets we diverage at this point and generate SPIR-V using the
   # llvm-spirv tool.
   if( ARG_ARCH STREQUAL spirv OR ARG_ARCH STREQUAL spirv64 )
-    set( obj_suffix ${ARG_ARCH_SUFFIX}.spv )
-    set( libclc_builtins_lib ${LIBCLC_OUTPUT_LIBRARY_DIR}/${obj_suffix} )
+    set( libclc_builtins_lib ${library_dir}/${LIBCLC_OUTPUT_FILENAME}.spv )
     if ( LIBCLC_USE_SPIRV_BACKEND )
       add_custom_command( OUTPUT ${libclc_builtins_lib}
-        COMMAND ${clang_exe} --target=${ARG_TRIPLE} -x ir -o ${libclc_builtins_lib} ${builtins_link_lib}
+        COMMAND ${clang_exe} -c --target=${ARG_TRIPLE} -x ir -o ${libclc_builtins_lib} ${builtins_link_lib}
         DEPENDS ${clang_target} ${builtins_link_lib} ${builtins_link_lib_tgt}
       )
     else()
@@ -419,8 +405,7 @@ function(add_libclc_builtin_set)
     endif()
   else()
     # Non-SPIR-V targets add an extra step to optimize the bytecode
-    set( obj_suffix ${ARG_ARCH_SUFFIX}.bc )
-    set( libclc_builtins_lib ${LIBCLC_OUTPUT_LIBRARY_DIR}/${obj_suffix} )
+    set( libclc_builtins_lib ${library_dir}/${LIBCLC_OUTPUT_FILENAME}.bc )
 
     add_custom_command( OUTPUT ${libclc_builtins_lib}
       COMMAND ${opt_exe} ${ARG_OPT_FLAGS} -o ${libclc_builtins_lib}
@@ -430,24 +415,28 @@ function(add_libclc_builtin_set)
   endif()
 
   # Add a 'library' target
-  add_custom_target( library-${obj_suffix} ALL DEPENDS ${libclc_builtins_lib} )
-  set_target_properties( "library-${obj_suffix}" PROPERTIES
+  add_custom_target( library-${ARG_ARCH_SUFFIX} ALL DEPENDS ${libclc_builtins_lib} )
+  set_target_properties( "library-${ARG_ARCH_SUFFIX}" PROPERTIES
     TARGET_FILE ${libclc_builtins_lib}
     FOLDER "libclc/Device IR/Library"
   )
 
-  # Also add a 'library' target for the triple. Since a triple may have
+  # Also add a 'libclc' target for the triple. Since a triple may have
   # multiple devices, ensure we only try to create the triple target once. The
   # triple's target will build all of the bytecode for its constituent devices.
-  if( NOT TARGET library-${ARG_TRIPLE} )
-    add_custom_target( library-${ARG_TRIPLE} ALL )
+  if( NOT TARGET libclc-${ARG_TRIPLE} )
+    add_custom_target( libclc-${ARG_TRIPLE} ALL )
   endif()
-  add_dependencies( library-${ARG_TRIPLE} library-${obj_suffix} )
+  add_dependencies( libclc-${ARG_TRIPLE} library-${ARG_ARCH_SUFFIX} )
   # Add dependency to top-level pseudo target to ease making other
   # targets dependent on libclc.
-  add_dependencies( ${ARG_PARENT_TARGET} library-${ARG_TRIPLE} )
+  add_dependencies( ${ARG_PARENT_TARGET} libclc-${ARG_TRIPLE} )
 
-  libclc_install(FILES ${libclc_builtins_lib})
+  # Install the created library.
+  install(
+    FILES ${libclc_builtins_lib}
+    DESTINATION ${LIBCLC_INSTALL_DIR}/${ARG_TRIPLE}
+  )
 
   # SPIR-V targets can exit early here
   if( ARG_ARCH STREQUAL spirv OR ARG_ARCH STREQUAL spirv64 )
@@ -460,7 +449,7 @@ function(add_libclc_builtin_set)
   # * nvptx64-- targets don't include workitem builtins
   # * clspv targets don't include all OpenCL builtins
   if( NOT ARG_ARCH MATCHES "^(nvptx|clspv)(64)?$" )
-    add_test( NAME external-funcs-${obj_suffix}
+    add_test( NAME external-funcs-${ARG_ARCH_SUFFIX}
       COMMAND ./check_external_funcs.sh ${libclc_builtins_lib} ${LLVM_TOOLS_BINARY_DIR}
       WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} )
   endif()
@@ -476,20 +465,26 @@ function(add_libclc_builtin_set)
       set(LIBCLC_LINK_OR_COPY copy)
     endif()
 
-    set( alias_suffix "${a}-${ARG_TRIPLE}.bc" )
+    file( MAKE_DIRECTORY ${LIBCLC_OUTPUT_LIBRARY_DIR}/${ARG_TRIPLE}/${a} )
+    set( libclc_alias_lib ${LIBCLC_OUTPUT_LIBRARY_DIR}/${ARG_TRIPLE}/${a}/${LIBCLC_OUTPUT_FILENAME}.bc )
     add_custom_command(
-      OUTPUT ${LIBCLC_OUTPUT_LIBRARY_DIR}/${alias_suffix}
-      COMMAND ${CMAKE_COMMAND} -E ${LIBCLC_LINK_OR_COPY} ${LIBCLC_LINK_OR_COPY_SOURCE} ${LIBCLC_OUTPUT_LIBRARY_DIR}/${alias_suffix}
-      DEPENDS library-${obj_suffix}
+      OUTPUT ${libclc_alias_lib}
+      COMMAND ${CMAKE_COMMAND} -E ${LIBCLC_LINK_OR_COPY} ${LIBCLC_LINK_OR_COPY_SOURCE} ${libclc_alias_lib}
+      DEPENDS library-${ARG_ARCH_SUFFIX}
     )
-    add_custom_target( alias-${alias_suffix} ALL
-      DEPENDS ${LIBCLC_OUTPUT_LIBRARY_DIR}/${alias_suffix}
+    add_custom_target( alias-${a}-${ARG_TRIPLE} ALL
+      DEPENDS ${libclc_alias_lib}
     )
-    add_dependencies( ${ARG_PARENT_TARGET} alias-${alias_suffix} )
-    set_target_properties( alias-${alias_suffix}
+    add_dependencies( ${ARG_PARENT_TARGET} alias-${a}-${ARG_TRIPLE} )
+    set_target_properties( alias-${a}-${ARG_TRIPLE}
       PROPERTIES FOLDER "libclc/Device IR/Aliases"
     )
-    libclc_install(FILES ${LIBCLC_OUTPUT_LIBRARY_DIR}/${alias_suffix})
+
+    # Install the library
+    install(
+      FILES ${libclc_alias_lib}
+      DESTINATION ${LIBCLC_INSTALL_DIR}/${ARG_TRIPLE}/${a}
+    )
   endforeach( a )
 endfunction(add_libclc_builtin_set)
 
diff --git a/libclc/opencl/include/clc/opencl/opencl-base.h b/libclc/opencl/include/clc/opencl/opencl-base.h
index ac0dd0f..78fb958 100644
--- a/libclc/opencl/include/clc/opencl/opencl-base.h
+++ b/libclc/opencl/include/clc/opencl/opencl-base.h
@@ -21,6 +21,5 @@
 #include <clc/clcfunc.h>
 
 /* 6.1 Supported Data Types */
-#include <clc/clctypes.h>
 
 #endif // __CLC_OPENCL_OPENCL_BASE_H__
diff --git a/libclc/opencl/include/clc/opencl/synchronization/utils.h b/libclc/opencl/include/clc/opencl/synchronization/utils.h
index 016341a..ec6401e 100644
--- a/libclc/opencl/include/clc/opencl/synchronization/utils.h
+++ b/libclc/opencl/include/clc/opencl/synchronization/utils.h
@@ -9,8 +9,8 @@
 #ifndef __CLC_OPENCL_SYNCHRONIZATION_UTILS_H__
 #define __CLC_OPENCL_SYNCHRONIZATION_UTILS_H__
 
-#include <clc/internal/clc.h>
 #include <clc/mem_fence/clc_mem_semantic.h>
+#include <clc/opencl/opencl-base.h>
 
 static _CLC_INLINE int __opencl_get_memory_scope(cl_mem_fence_flags flag) {
   if (flag & CLK_GLOBAL_MEM_FENCE)
diff --git a/libclc/opencl/include/clc/opencl/utils.h b/libclc/opencl/include/clc/opencl/utils.h
index 7304ce3..fefcecf 100644
--- a/libclc/opencl/include/clc/opencl/utils.h
+++ b/libclc/opencl/include/clc/opencl/utils.h
@@ -9,7 +9,7 @@
 #ifndef __CLC_OPENCL_UTILS_H__
 #define __CLC_OPENCL_UTILS_H__
 
-#include <clc/internal/clc.h>
+#include <clc/opencl/opencl-base.h>
 
 static _CLC_INLINE int __opencl_get_clang_memory_scope(memory_scope scope) {
   switch (scope) {