[libclc] Refine __clc_fp*_subnormals_supported

libclc/CMakeLists.txt

@@ -46,8 +46,6 @@ endif()
 set( LIBCLC_TARGETS_TO_BUILD ${LIBCLC_DEFAULT_TARGET}
     CACHE STRING "Semicolon-separated list of libclc targets to build, or 'all'." )
 
-option( ENABLE_RUNTIME_SUBNORMAL "Enable runtime linking of subnormal support." OFF )
-
 option(
   LIBCLC_USE_SPIRV_BACKEND "Build SPIR-V targets with the SPIR-V backend." OFF
 )
@@ -239,19 +237,6 @@ set( tahiti_aliases pitcairn verde oland hainan bonaire kabini kaveri hawaii
 configure_file( libclc.pc.in libclc.pc @ONLY )
 install( FILES ${CMAKE_CURRENT_BINARY_DIR}/libclc.pc DESTINATION "${CMAKE_INSTALL_DATADIR}/pkgconfig" )
 
-if( ENABLE_RUNTIME_SUBNORMAL )
-  foreach( file IN ITEMS subnormal_use_default subnormal_disable )
-    link_bc(
-       TARGET ${file}
-       INPUTS ${CMAKE_CURRENT_SOURCE_DIR}/opencl/lib/generic/${file}.ll
-    )
-    install(
-      FILES $<TARGET_PROPERTY:${file},TARGET_FILE>
-      DESTINATION "${CMAKE_INSTALL_DATADIR}/clc"
-    )
-  endforeach()
-endif()
-
 set_source_files_properties(
   # CLC builtins
   ${CMAKE_CURRENT_SOURCE_DIR}/clc/lib/generic/math/clc_native_cos.cl
@@ -350,12 +335,6 @@ foreach( t ${LIBCLC_TARGETS_TO_BUILD} )
 
   set( opencl_lib_files )
 
-  if( NOT ARCH STREQUAL spirv AND NOT ARCH STREQUAL spirv64 AND
-      NOT ARCH STREQUAL clspv AND NOT ARCH STREQUAL clspv64 AND
-      NOT ENABLE_RUNTIME_SUBNORMAL )
-    list( APPEND opencl_lib_files opencl/lib/generic/subnormal_use_default.ll )
-  endif()
-
   libclc_configure_lib_source(
     opencl_lib_files
     LIB_ROOT_DIR opencl
@@ -412,6 +391,7 @@ foreach( t ${LIBCLC_TARGETS_TO_BUILD} )
       # Error on undefined macros
       -Werror=undef
       -fdiscard-value-names
+      -fdenormal-fp-math=dynamic
     )
 
     if( NOT "${cpu}" STREQUAL "" )

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

@@ -10,7 +10,6 @@
 
 #include <clc/clcfunc.h>
 
-_CLC_DECL bool __clc_subnormals_disabled();
 _CLC_DECL bool __clc_fp16_subnormals_supported();
 _CLC_DECL bool __clc_fp32_subnormals_supported();
 _CLC_DECL bool __clc_fp64_subnormals_supported();

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

@@ -65,14 +65,18 @@ bool __attribute__((noinline)) __clc_runtime_has_hw_fma32(void);
 
 #define LOG_MAGIC_NUM_SP32 (1 + NUMEXPBITS_SP32 - EXPBIAS_SP32)
 
-_CLC_OVERLOAD _CLC_INLINE float __clc_flush_denormal_if_not_supported(float x) {
-  int ix = __clc_as_int(x);
-  if (!__clc_fp32_subnormals_supported() && ((ix & EXPBITS_SP32) == 0) &&
-      ((ix & MANTBITS_SP32) != 0)) {
-    ix &= SIGNBIT_SP32;
-    x = __clc_as_float(ix);
-  }
+static _CLC_INLINE float __clc_flush_denormal_if_not_supported(float x) {
+#ifdef CLC_SPIRV
+  // The function is only used in __clc_sw_fma.
+  // FIXME delete this once SPIR-V backend supports SPV_KHR_fma.
+  return __builtin_elementwise_abs(x) < 0x1p-149f
+             ? __builtin_elementwise_copysign(0.0f, x)
+             : x;
+#else
+  if (!__builtin_isfpclass(x, __FPCLASS_SNAN))
+    return __builtin_elementwise_canonicalize(x);
   return x;
+#endif
 }
 
 #ifdef cl_khr_fp64

libclc/clc/lib/generic/SOURCES

@@ -143,6 +143,7 @@ math/clc_sincos_helpers.cl
 math/clc_sinh.cl
 math/clc_sinpi.cl
 math/clc_sqrt.cl
+math/clc_subnormal_config.cl
 math/clc_sw_fma.cl
 math/clc_tables.cl
 math/clc_tan.cl

libclc/clc/lib/generic/math/clc_exp10.cl

@@ -11,7 +11,6 @@
 #include <clc/math/clc_fma.h>
 #include <clc/math/clc_ldexp.h>
 #include <clc/math/clc_mad.h>
-#include <clc/math/clc_subnormal_config.h>
 #include <clc/math/math.h>
 #include <clc/math/tables.h>
 #include <clc/relational/clc_isnan.h>

libclc/clc/lib/generic/math/clc_hypot.cl

@@ -12,7 +12,6 @@
 #include <clc/math/clc_fma.h>
 #include <clc/math/clc_mad.h>
 #include <clc/math/clc_sqrt.h>
-#include <clc/math/clc_subnormal_config.h>
 #include <clc/math/math.h>
 #include <clc/relational/clc_isnan.h>
 #include <clc/shared/clc_clamp.h>

libclc/clc/lib/generic/math/clc_pow.cl

@@ -12,7 +12,6 @@
 #include <clc/math/clc_fma.h>
 #include <clc/math/clc_ldexp.h>
 #include <clc/math/clc_mad.h>
-#include <clc/math/clc_subnormal_config.h>
 #include <clc/math/math.h>
 #include <clc/math/tables.h>
 #include <clc/relational/clc_select.h>

libclc/clc/lib/generic/math/clc_pown.cl

@@ -12,7 +12,6 @@
 #include <clc/math/clc_fma.h>
 #include <clc/math/clc_ldexp.h>
 #include <clc/math/clc_mad.h>
-#include <clc/math/clc_subnormal_config.h>
 #include <clc/math/math.h>
 #include <clc/math/tables.h>
 #include <clc/relational/clc_select.h>

libclc/clc/lib/generic/math/clc_powr.cl

@@ -12,7 +12,6 @@
 #include <clc/math/clc_fma.h>
 #include <clc/math/clc_ldexp.h>
 #include <clc/math/clc_mad.h>
-#include <clc/math/clc_subnormal_config.h>
 #include <clc/math/math.h>
 #include <clc/math/tables.h>
 #include <clc/relational/clc_select.h>

libclc/clc/lib/generic/math/clc_subnormal_config.cl

@@ -0,0 +1,44 @@
+//===----------------------------------------------------------------------===//
+//
+// 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/internal/clc.h>
+#include <clc/math/clc_fabs.h>
+#include <clc/math/clc_subnormal_config.h>
+
+#ifdef cl_khr_fp16
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+_CLC_DEF bool __clc_fp16_subnormals_supported() {
+#ifdef CLC_SPIRV
+  half x = __clc_fabs(0x1p-24h);
+#else
+  half x = __builtin_elementwise_canonicalize(0x1p-24h);
+#endif
+  return !__builtin_isfpclass(x, __FPCLASS_POSZERO);
+}
+#endif // cl_khr_fp16
+
+_CLC_DEF bool __clc_fp32_subnormals_supported() {
+#ifdef CLC_SPIRV
+  float x = __clc_fabs(0x1p-149f);
+#else
+  float x = __builtin_elementwise_canonicalize(0x1p-149f);
+#endif
+  return !__builtin_isfpclass(x, __FPCLASS_POSZERO);
+}
+
+#ifdef cl_khr_fp64
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+_CLC_DEF bool __clc_fp64_subnormals_supported() {
+#ifdef CLC_SPIRV
+  double x = __clc_fabs(0x1p-1074);
+#else
+  double x = __builtin_elementwise_canonicalize(0x1p-1074);
+#endif
+  return !__builtin_isfpclass(x, __FPCLASS_POSZERO);
+}
+#endif // cl_khr_fp64

libclc/opencl/lib/clspv/SOURCES

@@ -4,7 +4,6 @@ conversion/convert_int2float.cl
 conversion/convert_integer.cl
 math/fma.cl
 shared/vstore_half.cl
-subnormal_config.cl
 ../generic/geometric/distance.cl
 ../generic/geometric/length.cl
 ../generic/math/acos.cl

libclc/opencl/lib/generic/SOURCES

@@ -1,5 +1,3 @@
-subnormal_config.cl
-subnormal_helper_func.ll
 async/async_work_group_copy.cl
 async/async_work_group_strided_copy.cl
 async/prefetch.cl

libclc/opencl/lib/spirv/SOURCES

@@ -1,4 +1,3 @@
-subnormal_config.cl
 ../generic/async/async_work_group_strided_copy.cl
 ../generic/async/wait_group_events.cl
 ../generic/common/degrees.cl