[compiler-rt] Move up undefined macro checks

[abrachet]

Summary

Previously HWCAP_ATOMIC and others were being used before checking if
they were defined. This moves up all the ifndef checks to define these
macros if they are not yet defined.

[ilinpv]

LGTM, thanks for fixing this!

[dim]

What was the original reason for this particular change? Because after this, with clang and compiler-rt 16, I now see a whole bunch of:

In file included from /home/dim/src/freebsd/llvm-16-update/contrib/llvm-project/compiler-rt/lib/builtins/cpu_model.c:1019:
In file included from /home/dim/obj/home/dim/src/freebsd/llvm-16-update/arm64.aarch64/tmp/usr/include/sys/auxv.h:35:
In file included from /home/dim/obj/home/dim/src/freebsd/llvm-16-update/arm64.aarch64/tmp/usr/include/machine/elf.h:36:
In file included from /home/dim/obj/home/dim/src/freebsd/llvm-16-update/arm64.aarch64/tmp/usr/include/sys/elf32.h:34:
/home/dim/obj/home/dim/src/freebsd/llvm-16-update/arm64.aarch64/tmp/usr/include/sys/elf_common.h:980:9: error: 'AT_HWCAP' macro redefined [-Werror,-Wmacro-redefined]
#define AT_HWCAP        25      /* CPU feature flags. */
        ^
/home/dim/src/freebsd/llvm-16-update/contrib/llvm-project/compiler-rt/lib/builtins/cpu_model.c:843:9: note: previous definition is here
#define AT_HWCAP 16
        ^
In file included from /home/dim/src/freebsd/llvm-16-update/contrib/llvm-project/compiler-rt/lib/builtins/cpu_model.c:1019:
In file included from /home/dim/obj/home/dim/src/freebsd/llvm-16-update/arm64.aarch64/tmp/usr/include/sys/auxv.h:35:
/home/dim/obj/home/dim/src/freebsd/llvm-16-update/arm64.aarch64/tmp/usr/include/machine/elf.h:95:9: error: 'HWCAP_FP' macro redefined [-Werror,-Wmacro-redefined]
#define HWCAP_FP                0x00000001
        ^
/home/dim/src/freebsd/llvm-16-update/contrib/llvm-project/compiler-rt/lib/builtins/cpu_model.c:849:9: note: previous definition is here
#define HWCAP_FP (1 << 0)
        ^
In file included from /home/dim/src/freebsd/llvm-16-update/contrib/llvm-project/compiler-rt/lib/builtins/cpu_model.c:1019:
In file included from /home/dim/obj/home/dim/src/freebsd/llvm-16-update/arm64.aarch64/tmp/usr/include/sys/auxv.h:35:
/home/dim/obj/home/dim/src/freebsd/llvm-16-update/arm64.aarch64/tmp/usr/include/machine/elf.h:96:9: error: 'HWCAP_ASIMD' macro redefined [-Werror,-Wmacro-redefined]
#define HWCAP_ASIMD             0x00000002
        ^
...

because it seems to me that checking whether a macro is defined should be done *after* including any system header where such a macro _could_ possibly be defined, not the other way around?

I think the order should instead be:

1. include system headers that potentially defined HWCAP_xxx
2. block with #ifndef HWCAP_xxx ... #define HWCAP_xxx ...
3. block with inline functions actually using HWCAP_xxx

[ilinpv]

I think the order should instead be:

1. include system headers that potentially defined HWCAP_xxx
2. block with #ifndef HWCAP_xxx ... #define HWCAP_xxx ...
3. block with inline functions actually using HWCAP_xxx

Absolutely correct. I think system headers includes

#if defined(__has_include)
#if __has_include(<sys/auxv.h>)
#include <sys/auxv.h>
#if __has_include(<asm/hwcap.h>)
#include <asm/hwcap.h>

#if defined(__ANDROID__)
#include <string.h>
#include <sys/system_properties.h>
#elif defined(__Fuchsia__)
#include <zircon/features.h>
#include <zircon/syscalls.h>
#endif

was moved after #ifndef HWCAP_xxx by mistake. Originally AT_HWCAP and HWCAP_ATOMICS were used before #ifndef HWCAP_xxx ... #define HWCAP_xxx ...

compiler-rt/lib/builtins/cpu_model.c

int CONSTRUCTOR_ATTRIBUTE __cpu_indicator_init(void) {
 
   assert(__cpu_model.__cpu_vendor < VENDOR_MAX);
   assert(__cpu_model.__cpu_type < CPU_TYPE_MAX);
   assert(__cpu_model.__cpu_subtype < CPU_SUBTYPE_MAX);
 
   return 0;
 }
 #elif defined(__aarch64__)
-// LSE support detection for out-of-line atomics
-// using HWCAP and Auxiliary vector
-_Bool __aarch64_have_lse_atomics
-    __attribute__((visibility("hidden"), nocommon));
-
-#if defined(__has_include)
-#if __has_include(<sys/auxv.h>)
-#include <sys/auxv.h>
-#if __has_include(<asm/hwcap.h>)
-#include <asm/hwcap.h>
-
-#if defined(__ANDROID__)
-#include <string.h>
-#include <sys/system_properties.h>
-#elif defined(__Fuchsia__)
-#include <zircon/features.h>
-#include <zircon/syscalls.h>
-#endif
-
-// Detect Exynos 9810 CPU
-#define IF_EXYNOS9810                                                          \
-  char arch[PROP_VALUE_MAX];                                                   \
-  if (__system_property_get("ro.arch", arch) > 0 &&                            \
-      strncmp(arch, "exynos9810", sizeof("exynos9810") - 1) == 0)
-
-static void CONSTRUCTOR_ATTRIBUTE init_have_lse_atomics(void) {
-#if defined(__FreeBSD__)
-  unsigned long hwcap;
-  int result = elf_aux_info(AT_HWCAP, &hwcap, sizeof hwcap);
-  __aarch64_have_lse_atomics = result == 0 && (hwcap & HWCAP_ATOMICS) != 0;
-#elif defined(__Fuchsia__)
-  // This ensures the vDSO is a direct link-time dependency of anything that
-  // needs this initializer code.
-#pragma comment(lib, "zircon")
-  uint32_t features;
-  zx_status_t status = _zx_system_get_features(ZX_FEATURE_KIND_CPU, &features);
-  __aarch64_have_lse_atomics =
-      status == ZX_OK && (features & ZX_ARM64_FEATURE_ISA_ATOMICS) != 0;
-#else
-  unsigned long hwcap = getauxval(AT_HWCAP);
-  _Bool result = (hwcap & HWCAP_ATOMICS) != 0;
-#if defined(__ANDROID__)
-  if (result) {
-    // Some cores in the Exynos 9810 CPU are ARMv8.2 and others are ARMv8.0;
-    // only the former support LSE atomics.  However, the kernel in the
-    // initial Android 8.0 release of Galaxy S9/S9+ devices incorrectly
-    // reported the feature as being supported.
-    //
-    // The kernel appears to have been corrected to mark it unsupported as of
-    // the Android 9.0 release on those devices, and this issue has not been
-    // observed anywhere else. Thus, this workaround may be removed if
-    // compiler-rt ever drops support for Android 8.0.
-    IF_EXYNOS9810 result = false;
-  }
-#endif // defined(__ANDROID__)
-  __aarch64_have_lse_atomics = result;
-#endif // defined(__FreeBSD__)
-}
-
-#if !defined(DISABLE_AARCH64_FMV)
-// CPUFeatures must correspond to the same AArch64 features in
-// AArch64TargetParser.h
-enum CPUFeatures {
-  FEAT_RNG,
-  FEAT_FLAGM,
-  FEAT_FLAGM2,
-  FEAT_FP16FML,
-  FEAT_DOTPROD,
-  FEAT_SM4,
-  FEAT_RDM,
-  FEAT_LSE,
-  FEAT_FP,
-  FEAT_SIMD,
-  FEAT_CRC,
-  FEAT_SHA1,
-  FEAT_SHA2,
-  FEAT_SHA3,
-  FEAT_AES,
-  FEAT_PMULL,
-  FEAT_FP16,
-  FEAT_DIT,
-  FEAT_DPB,
-  FEAT_DPB2,
-  FEAT_JSCVT,
-  FEAT_FCMA,
-  FEAT_RCPC,
-  FEAT_RCPC2,
-  FEAT_FRINTTS,
-  FEAT_DGH,
-  FEAT_I8MM,
-  FEAT_BF16,
-  FEAT_EBF16,
-  FEAT_RPRES,
-  FEAT_SVE,
-  FEAT_SVE_BF16,
-  FEAT_SVE_EBF16,
-  FEAT_SVE_I8MM,
-  FEAT_SVE_F32MM,
-  FEAT_SVE_F64MM,
-  FEAT_SVE2,
-  FEAT_SVE_AES,
-  FEAT_SVE_PMULL128,
-  FEAT_SVE_BITPERM,
-  FEAT_SVE_SHA3,
-  FEAT_SVE_SM4,
-  FEAT_SME,
-  FEAT_MEMTAG,
-  FEAT_MEMTAG2,
-  FEAT_MEMTAG3,
-  FEAT_SB,
-  FEAT_PREDRES,
-  FEAT_SSBS,
-  FEAT_SSBS2,
-  FEAT_BTI,
-  FEAT_LS64,
-  FEAT_LS64_V,
-  FEAT_LS64_ACCDATA,
-  FEAT_WFXT,
-  FEAT_SME_F64,
-  FEAT_SME_I64,
-  FEAT_SME2,
-  FEAT_MAX
-};
-
-// Architecture features used
-// in Function Multi Versioning
-struct {
-  unsigned long long features;
-  // As features grows new fields could be added
-} __aarch64_cpu_features __attribute__((visibility("hidden"), nocommon));
 
 #ifndef AT_HWCAP
 #define AT_HWCAP 16
 #endif
 #ifndef HWCAP_CPUID
 #define HWCAP_CPUID (1 << 11)
 #endif
 #ifndef HWCAP_FP

 #endif
 #ifndef HWCAP2_EBF16
 #define HWCAP2_EBF16 (1UL << 32)
 #endif
 #ifndef HWCAP2_SVE_EBF16
 #define HWCAP2_SVE_EBF16 (1UL << 33)
 #endif
 
+// LSE support detection for out-of-line atomics
+// using HWCAP and Auxiliary vector
+_Bool __aarch64_have_lse_atomics
+    __attribute__((visibility("hidden"), nocommon));
+
+#if defined(__has_include)
+#if __has_include(<sys/auxv.h>)
+#include <sys/auxv.h>
+#if __has_include(<asm/hwcap.h>)
+#include <asm/hwcap.h>
+
+#if defined(__ANDROID__)
+#include <string.h>
+#include <sys/system_properties.h>
+#elif defined(__Fuchsia__)
+#include <zircon/features.h>
+#include <zircon/syscalls.h>
+#endif
+
+// Detect Exynos 9810 CPU
+#define IF_EXYNOS9810                                                          \
+  char arch[PROP_VALUE_MAX];                                                   \
+  if (__system_property_get("ro.arch", arch) > 0 &&                            \
+      strncmp(arch, "exynos9810", sizeof("exynos9810") - 1) == 0)
+
+static void CONSTRUCTOR_ATTRIBUTE init_have_lse_atomics(void) {
+#if defined(__FreeBSD__)
+  unsigned long hwcap;
+  int result = elf_aux_info(AT_HWCAP, &hwcap, sizeof hwcap);
+  __aarch64_have_lse_atomics = result == 0 && (hwcap & HWCAP_ATOMICS) != 0;
+#elif defined(__Fuchsia__)
+  // This ensures the vDSO is a direct link-time dependency of anything that
+  // needs this initializer code.
+#pragma comment(lib, "zircon")
+  uint32_t features;
+  zx_status_t status = _zx_system_get_features(ZX_FEATURE_KIND_CPU, &features);
+  __aarch64_have_lse_atomics =
+      status == ZX_OK && (features & ZX_ARM64_FEATURE_ISA_ATOMICS) != 0;
+#else
+  unsigned long hwcap = getauxval(AT_HWCAP);
+  _Bool result = (hwcap & HWCAP_ATOMICS) != 0;
+#if defined(__ANDROID__)
+  if (result) {
+    // Some cores in the Exynos 9810 CPU are ARMv8.2 and others are ARMv8.0;
+    // only the former support LSE atomics.  However, the kernel in the
+    // initial Android 8.0 release of Galaxy S9/S9+ devices incorrectly
+    // reported the feature as being supported.
+    //
+    // The kernel appears to have been corrected to mark it unsupported as of
+    // the Android 9.0 release on those devices, and this issue has not been
+    // observed anywhere else. Thus, this workaround may be removed if
+    // compiler-rt ever drops support for Android 8.0.
+    IF_EXYNOS9810 result = false;
+  }
+#endif // defined(__ANDROID__)
+  __aarch64_have_lse_atomics = result;
+#endif // defined(__FreeBSD__)
+}
+
+#if !defined(DISABLE_AARCH64_FMV)
+// CPUFeatures must correspond to the same AArch64 features in
+// AArch64TargetParser.h
+enum CPUFeatures {
+  FEAT_RNG,
+  FEAT_FLAGM,
+  FEAT_FLAGM2,
+  FEAT_FP16FML,
+  FEAT_DOTPROD,
+  FEAT_SM4,
+  FEAT_RDM,
+  FEAT_LSE,
+  FEAT_FP,
+  FEAT_SIMD,
+  FEAT_CRC,
+  FEAT_SHA1,
+  FEAT_SHA2,
+  FEAT_SHA3,
+  FEAT_AES,
+  FEAT_PMULL,
+  FEAT_FP16,
+  FEAT_DIT,
+  FEAT_DPB,
+  FEAT_DPB2,
+  FEAT_JSCVT,
+  FEAT_FCMA,
+  FEAT_RCPC,
+  FEAT_RCPC2,
+  FEAT_FRINTTS,
+  FEAT_DGH,
+  FEAT_I8MM,
+  FEAT_BF16,
+  FEAT_EBF16,
+  FEAT_RPRES,
+  FEAT_SVE,
+  FEAT_SVE_BF16,
+  FEAT_SVE_EBF16,
+  FEAT_SVE_I8MM,
+  FEAT_SVE_F32MM,
+  FEAT_SVE_F64MM,
+  FEAT_SVE2,
+  FEAT_SVE_AES,
+  FEAT_SVE_PMULL128,
+  FEAT_SVE_BITPERM,
+  FEAT_SVE_SHA3,
+  FEAT_SVE_SM4,
+  FEAT_SME,
+  FEAT_MEMTAG,
+  FEAT_MEMTAG2,
+  FEAT_MEMTAG3,
+  FEAT_SB,
+  FEAT_PREDRES,
+  FEAT_SSBS,
+  FEAT_SSBS2,
+  FEAT_BTI,
+  FEAT_LS64,
+  FEAT_LS64_V,
+  FEAT_LS64_ACCDATA,
+  FEAT_WFXT,
+  FEAT_SME_F64,
+  FEAT_SME_I64,
+  FEAT_SME2,
+  FEAT_MAX
+};
+
+// Architecture features used
+// in Function Multi Versioning
+struct {
+  unsigned long long features;
+  // As features grows new fields could be added
+} __aarch64_cpu_features __attribute__((visibility("hidden"), nocommon));
+
 void init_cpu_features_resolver(unsigned long hwcap, unsigned long hwcap2) {
 #define setCPUFeature(F) __aarch64_cpu_features.features |= 1ULL << F
 #define getCPUFeature(id, ftr) __asm__("mrs %0, " #id : "=r"(ftr))
 #define extractBits(val, start, number)                                        \
   (val & ((1ULL << number) - 1ULL) << start) >> start
   if (hwcap & HWCAP_CRC32)
     setCPUFeature(FEAT_CRC);
   if (hwcap & HWCAP_PMULL)