[compiler-rt][ARM] Optimized mulsf3 and divsf3 (#168394)

(Reland of #161546, fixing three build and test issues)

This commit adds optimized assembly versions of single-precision float
multiplication and division. Both functions are implemented in a style
that can be assembled as either of Arm and Thumb2; for multiplication, a
separate implementation is provided for Thumb1. Also, extensive new
tests are added for multiplication and division.

These implementations can be removed from the build by defining the
cmake variable COMPILER_RT_ARM_OPTIMIZED_FP=OFF.

Outlying parts of the functionality which are not on the fast path, such
as NaN handling and underflow, are handled in helper functions written
in C. These can be shared between the Arm/Thumb2 and Thumb1
implementations, and also reused by other optimized assembly functions
we hope to add in future.

Author: statham-arm

compiler-rt/cmake/Modules/CheckAssemblerFlag.cmake

@@ -0,0 +1,39 @@
+# Helper function to find out whether the assembler supports a particular
+# command-line flag. You'd like to use the standard check_compiler_flag(), but
+# that only supports a fixed list of languages, and ASM isn't one of them. So
+# we do it ourselves, by trying to assemble an empty source file.
+
+function(check_assembler_flag outvar flag)
+  if(NOT DEFINED "${outvar}")
+    if(NOT CMAKE_REQUIRED_QUIET)
+      message(CHECK_START "Checking for assembler flag ${flag}")
+    endif()
+
+    # Stop try_compile from attempting to link the result of the assembly, so
+    # that we don't depend on having a working linker, and also don't have to
+    # figure out what special symbol like _start needs to be defined in the
+    # test input.
+    #
+    # This change is made within the dynamic scope of this function, so
+    # CMAKE_TRY_COMPILE_TARGET_TYPE will be restored to its previous value on
+    # return.
+    set(CMAKE_TRY_COMPILE_TARGET_TYPE STATIC_LIBRARY)
+
+    # Try to assemble an empty file with a .S name, using the provided flag.
+    set(asm_source_file
+      ${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CheckAssemblerFlag.S)
+    write_file(${asm_source_file} "")
+    try_compile(${outvar}
+      ${CMAKE_BINARY_DIR}
+      SOURCES ${asm_source_file}
+      COMPILE_DEFINITIONS ${flag})
+
+    if(NOT CMAKE_REQUIRED_QUIET)
+      if(${outvar})
+        message(CHECK_PASS "Accepted")
+      else()
+        message(CHECK_FAIL "Not accepted")
+      endif()
+    endif()
+  endif()
+endfunction()

compiler-rt/lib/builtins/CMakeLists.txt

@@ -60,6 +60,7 @@ endif()
 include(builtin-config-ix)
 include(CMakeDependentOption)
 include(CMakePushCheckState)
+include(CheckAssemblerFlag)
 
 option(COMPILER_RT_BUILTINS_HIDE_SYMBOLS
   "Do not export any symbols from the static library." ON)
@@ -423,6 +424,40 @@ set(arm_or_thumb2_base_SOURCES
   ${GENERIC_SOURCES}
 )
 
+option(COMPILER_RT_ARM_OPTIMIZED_FP
+  "On 32-bit Arm, use optimized assembly implementations of FP arithmetic. Likely to increase code size, but be faster." ON)
+
+set(arm_or_thumb2_optimized_fp_SOURCES)
+if(COMPILER_RT_ARM_OPTIMIZED_FP AND BUILTIN_SUPPORTED_ARCH MATCHES "arm")
+  check_assembler_flag(COMPILER_RT_HAS_MIMPLICIT_IT -mimplicit-it=always)
+  if(COMPILER_RT_HAS_MIMPLICIT_IT)
+    set(implicit_it_flag -mimplicit-it=always)
+  else()
+    check_assembler_flag(
+      COMPILER_RT_HAS_WA_MIMPLICIT_IT -Wa,-mimplicit-it=always)
+    if(COMPILER_RT_HAS_WA_MIMPLICIT_IT)
+      set(implicit_it_flag -Wa,-mimplicit-it=always)
+    else()
+      message(WARNING "Don't know how to set the -mimplicit-it=always flag in this assembler; not including Arm optimized implementations")
+      set(implicit_it_flag "")
+    endif()
+  endif()
+
+  if(implicit_it_flag)
+    set(assembly_files
+      arm/mulsf3.S
+      arm/divsf3.S)
+    set_source_files_properties(${assembly_files}
+      PROPERTIES COMPILE_OPTIONS ${implicit_it_flag})
+    set(arm_or_thumb2_optimized_fp_SOURCES
+      ${assembly_files}
+      arm/fnan2.c
+      arm/fnorm2.c
+      arm/funder.c
+      )
+  endif()
+endif()
+
 set(arm_sync_SOURCES
   arm/sync_fetch_and_add_4.S
   arm/sync_fetch_and_add_8.S
@@ -456,6 +491,16 @@ set(thumb1_base_SOURCES
   ${GENERIC_SOURCES}
 )
 
+if(COMPILER_RT_ARM_OPTIMIZED_FP)
+  set(thumb1_base_SOURCES
+    arm/thumb1/mulsf3.S
+    arm/fnan2.c
+    arm/fnorm2.c
+    arm/funder.c
+    ${thumb1_base_SOURCES}
+  )
+endif()
+
 set(arm_EABI_RT_SOURCES
   arm/aeabi_cdcmp.S
   arm/aeabi_cdcmpeq_check_nan.c
@@ -567,6 +612,7 @@ if(MINGW)
     arm/aeabi_uldivmod.S
     arm/chkstk.S
     ${arm_or_thumb2_base_SOURCES}
+    ${arm_or_thumb2_optimized_fp_SOURCES}
     ${arm_sync_SOURCES}
   )
 
@@ -577,6 +623,7 @@ elseif(NOT WIN32)
   # TODO the EABI sources should only be added to EABI targets
   set(arm_SOURCES
     ${arm_or_thumb2_base_SOURCES}
+    ${arm_or_thumb2_optimized_fp_SOURCES}
     ${arm_sync_SOURCES}
     ${arm_EABI_SOURCES}
     ${arm_Thumb1_SOURCES}