Preparation for Power optimizations

Optimized functions for Power will make use of GNU indirect functions,
an extension to support different implementations of the same function,
which can be selected during runtime. This will be used to provide
optimized functions for different processor versions.

Since this is a GNU extension, we placed the definition of the Z_IFUNC
macro under `contrib/gcc`. This can be reused by other archs as well.

Author: Matheus Castanho <msc@linux.ibm.com>
Author: Rogerio Alves <rcardoso@linux.ibm.com>

CMakeLists.txt

@@ -7,6 +7,7 @@ set(VERSION "1.2.12.1")
 
 option(ASM686 "Enable building i686 assembly implementation")
 option(AMD64 "Enable building amd64 assembly implementation")
+option(POWER "Enable building power implementation")
 
 set(INSTALL_BIN_DIR "${CMAKE_INSTALL_PREFIX}/bin" CACHE PATH "Installation directory for executables")
 set(INSTALL_LIB_DIR "${CMAKE_INSTALL_PREFIX}/lib" CACHE PATH "Installation directory for libraries")
@@ -140,6 +141,72 @@ if(CMAKE_COMPILER_IS_GNUCC)
 		add_definitions(-DASMV)
 		set_source_files_properties(${ZLIB_ASMS} PROPERTIES LANGUAGE C COMPILE_FLAGS -DNO_UNDERLINE)
 	endif()
+
+    # test to see if we can use a GNU indirect function to detect and load optimized code at runtime
+    CHECK_C_SOURCE_COMPILES("
+    static int test_ifunc_native(void)
+    {
+       return 1;
+    }
+    static int (*(check_ifunc_native(void)))(void)
+    {
+       return test_ifunc_native;
+    }
+    int test_ifunc(void) __attribute__ ((ifunc (\"check_ifunc_native\")));
+    int main(void)
+    {
+       return 0;
+    }
+    " HAS_C_ATTR_IFUNC)
+
+    if(HAS_C_ATTR_IFUNC)
+        add_definitions(-DHAVE_IFUNC)
+        set(ZLIB_PRIVATE_HDRS ${ZLIB_PRIVATE_HDRS} contrib/gcc/zifunc.h)
+    endif()
+
+    if(POWER)
+        # Test to see if we can use the optimizations for Power
+        CHECK_C_SOURCE_COMPILES("
+        #ifndef _ARCH_PPC
+            #error \"Target is not Power\"
+        #endif
+        #ifndef __BUILTIN_CPU_SUPPORTS__
+            #error \"Target doesn't support __builtin_cpu_supports()\"
+        #endif
+        int main() { return 0; }
+        " HAS_POWER_SUPPORT)
+
+        if(HAS_POWER_SUPPORT AND HAS_C_ATTR_IFUNC)
+            add_definitions(-DZ_POWER_OPT)
+
+            set(CMAKE_REQUIRED_FLAGS -mcpu=power8)
+            CHECK_C_SOURCE_COMPILES("int main(void){return 0;}" POWER8)
+
+            if(POWER8)
+                add_definitions(-DZ_POWER8)
+                set(ZLIB_POWER8 )
+
+                set_source_files_properties(
+                    ${ZLIB_POWER8}
+                    PROPERTIES COMPILE_FLAGS -mcpu=power8)
+            endif()
+
+            set(CMAKE_REQUIRED_FLAGS -mcpu=power9)
+            CHECK_C_SOURCE_COMPILES("int main(void){return 0;}" POWER9)
+
+            if(POWER9)
+                add_definitions(-DZ_POWER9)
+                set(ZLIB_POWER9 )
+
+                set_source_files_properties(
+                    ${ZLIB_POWER9}
+                    PROPERTIES COMPILE_FLAGS -mcpu=power9)
+            endif()
+
+            set(ZLIB_PRIVATE_HDRS ${ZLIB_PRIVATE_HDRS} contrib/power/power.h)
+            set(ZLIB_SRCS ${ZLIB_SRCS} ${ZLIB_POWER8} ${ZLIB_POWER9})
+        endif()
+    endif()
 endif()
 
 if(MSVC)

configure

@@ -836,6 +836,72 @@ EOF
   fi
 fi
 
+# test to see if we can use a gnu indirection function to detect and load optimized code at runtime
+echo >> configure.log
+cat > $test.c <<EOF
+static int test_ifunc_native(void)
+{
+  return 1;
+}
+
+static int (*(check_ifunc_native(void)))(void)
+{
+  return test_ifunc_native;
+}
+
+int test_ifunc(void) __attribute__ ((ifunc ("check_ifunc_native")));
+EOF
+
+if tryboth $CC -c $CFLAGS $test.c; then
+  SFLAGS="${SFLAGS} -DHAVE_IFUNC"
+  CFLAGS="${CFLAGS} -DHAVE_IFUNC"
+  echo "Checking for attribute(ifunc) support... Yes." | tee -a configure.log
+else
+  echo "Checking for attribute(ifunc) support... No." | tee -a configure.log
+fi
+
+# Test to see if we can use the optimizations for Power
+echo >> configure.log
+cat > $test.c <<EOF
+#ifndef _ARCH_PPC
+  #error "Target is not Power"
+#endif
+#ifndef HAVE_IFUNC
+  #error "Target doesn't support ifunc"
+#endif
+#ifndef __BUILTIN_CPU_SUPPORTS__
+  #error "Target doesn't support __builtin_cpu_supports()"
+#endif
+EOF
+
+if tryboth $CC -c $CFLAGS $test.c; then
+  echo "int main(void){return 0;}" > $test.c
+
+  if tryboth $CC -c $CFLAGS -mcpu=power8 $test.c; then
+    POWER8="-DZ_POWER8"
+    PIC_OBJC="${PIC_OBJC}"
+    OBJC="${OBJC}"
+    echo "Checking for -mcpu=power8 support... Yes." | tee -a configure.log
+  else
+    echo "Checking for -mcpu=power8 support... No." | tee -a configure.log
+  fi
+
+  if tryboth $CC -c $CFLAGS -mcpu=power9 $test.c; then
+    POWER9="-DZ_POWER9"
+    PIC_OBJC="${PIC_OBJC}"
+    OBJC="${OBJC}"
+    echo "Checking for -mcpu=power9 support... Yes." | tee -a configure.log
+  else
+    echo "Checking for -mcpu=power9 support... No." | tee -a configure.log
+  fi
+
+  SFLAGS="${SFLAGS} ${POWER8} ${POWER9} -DZ_POWER_OPT"
+  CFLAGS="${CFLAGS} ${POWER8} ${POWER9} -DZ_POWER_OPT"
+  echo "Checking for Power optimizations support... Yes." | tee -a configure.log
+else
+  echo "Checking for Power optimizations support... No." | tee -a configure.log
+fi
+
 # show the results in the log
 echo >> configure.log
 echo ALL = $ALL >> configure.log

contrib/README.contrib

@@ -11,6 +11,10 @@ ada/        by Dmitriy Anisimkov <anisimkov@yahoo.com>
 blast/      by Mark Adler <madler@alumni.caltech.edu>
         Decompressor for output of PKWare Data Compression Library (DCL)
 
+gcc/        by Matheus Castanho <msc@linux.ibm.com>
+            and Rogerio Alves <rcardoso@linux.ibm.com>
+        Optimization helpers using GCC-specific extensions
+
 delphi/     by Cosmin Truta <cosmint@cs.ubbcluj.ro>
         Support for Delphi and C++ Builder
 
@@ -42,6 +46,10 @@ minizip/    by Gilles Vollant <info@winimage.com>
 pascal/     by Bob Dellaca <bobdl@xtra.co.nz> et al.
         Support for Pascal
 
+power/      by Matheus Castanho <msc@linux.ibm.com>
+            and Rogerio Alves <rcardoso@linux.ibm.com>
+        Optimized functions for Power processors
+
 puff/       by Mark Adler <madler@alumni.caltech.edu>
         Small, low memory usage inflate.  Also serves to provide an
         unambiguous description of the deflate format.

contrib/gcc/zifunc.h

@@ -0,0 +1,60 @@
+/* Copyright (C) 2019 Matheus Castanho <msc@linux.ibm.com>, IBM
+ *               2019 Rogerio Alves    <rogerio.alves@ibm.com>, IBM
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ */
+
+#ifndef Z_IFUNC_H_
+#define Z_IFUNC_H_
+
+/* Helpers for arch optimizations */
+
+#define Z_IFUNC(fname) \
+    typeof(fname) fname __attribute__ ((ifunc (#fname "_resolver"))); \
+    local typeof(fname) *fname##_resolver(void)
+/* This is a helper macro to declare a resolver for an indirect function
+ * (ifunc). Let's say you have function
+ *
+ *    int foo (int a);
+ *
+ * for which you want to provide different implementations, for example:
+ *
+ *    int foo_clever (int a) {
+ *      ... clever things ...
+ *    }
+ *
+ *    int foo_smart (int a) {
+ *      ... smart things ...
+ *    }
+ *
+ * You will have to declare foo() as an indirect function and also provide a
+ * resolver for it, to choose between foo_clever() and foo_smart() based on
+ * some criteria you define (e.g. processor features).
+ *
+ * Since most likely foo() has a default implementation somewhere in zlib, you
+ * may have to rename it so the 'foo' symbol can be used by the ifunc without
+ * conflicts.
+ *
+ *    #define foo foo_default
+ *    int foo (int a) {
+ *      ...
+ *    }
+ *    #undef foo
+ *
+ * Now you just have to provide a resolver function to choose which function
+ * should be used (decided at runtime on the first call to foo()):
+ *
+ *    Z_IFUNC(foo) {
+ *        if (... some condition ...)
+ *          return foo_clever;
+ *
+ *        if (... other condition ...)
+ *          return foo_smart;
+ *
+ *        return foo_default;
+ *    }
+ *
+ * All calls to foo() throughout the code can remain untouched, all the magic
+ * will be done by the linker using the resolver function.
+ */
+
+#endif /* Z_IFUNC_H_ */

contrib/power/power.h

@@ -0,0 +1,4 @@
+/* Copyright (C) 2019 Matheus Castanho <msc@linux.ibm.com>, IBM
+ *               2019 Rogerio Alves    <rogerio.alves@ibm.com>, IBM
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ */