Emit the ARM build attributes ABI_PCS_wchar_t and ABI_enum_size.

[olista01]

Summary

Emit the ARM build attributes ABI_PCS_wchar_t and ABI_enum_size based on module flags metadata.

[rengolin]

include/llvm/Support/ARMBuildAttributes.h

163

you're never using, nor testing this...

175

you're never using, nor testing this...

180

you're never using, nor testing this...

test/CodeGen/ARM/metadata-default.ll

14

shouldn't this value be the actual number of bytes used? so than you can forbid usage from the front-end later.

15

I think this one could be a bit more expressive, such as using a string, so that we can disable or use the fourth style, which is currently inaccessible.

[compnerd]

Some minor nits, but LGTM. Thanks for adding support for this EABI attribute.

include/llvm/Support/ARMBuildAttributes.h

162

Can you add the uleb128 encoding and type similar to the other enums please?

174

Same here.

lib/Target/ARM/ARMAsmPrinter.cpp

736

You should be able to collapse the cast and if into a single statement.

I think you can use auto here without loss of clarity:

if (auto WCharType = cast_or_null<ConstantInt>(SM->getModuleFlag("short_wchar"))

746

Similarly here.

test/CodeGen/ARM/metadata-default.ll

5

The Module ID doesnt add much to the test. You can trim the test to just the RUN line, the define, and metadata.

[olista01]

you're never using, nor testing this... x3

I was following the example of the other build attributes (e.g. the position independent code ones), including all values specified in the ABI addenda, even if we do not use them.

[olista01]

Changes corresponding to matching change in D4212 (store sizes of wchar_t and enums in metadata, not just boolean flags)

[rengolin]

Hi Oliver,

I'm happy with this patch, but we need to make sure we don't create those attributes on other architectures in D4212.

We can overlook the non use of the extra options (as you said, following current implementation), but as it stands, it's not just unused, but impossible to use from a language point of view. That's why I proposed a textual format (for the ABI type), but this can be done later. Please add a FIXME to that regard.

cheers,
--renato

[olista01]

Thanks, committed revision 211349.

docs/LangRef.rst

 strings to specify e.g., a single library, while still having that specifier be
 preserved as an atomic element that can be recognized by a target specific
 assembly writer or object file emitter.
 
 Each individual option is required to be either a valid option for the target's
 linker, or an option that is reserved by the target specific assembly writer or
 object file emitter. No other aspect of these options is defined by the IR.
 
+C type width Module Flags Metadata
+----------------------------------
+
+The ARM backend emits a section into each generated object file describing the
+options that it was compiled with (in a compiler-independent way) to prevent
+linking incompatible objects, and to allow automatic library selection. Some
+of these options are not visible at the IR level, namely wchar_t width and enum
+width.
+
+To pass this information to the backend, these options are encoded in module
+flags metadata, using the following key-value pairs:
+
+.. list-table::
+   :header-rows: 1
+   :widths: 30 70
+
+   * - Key
+     - Value
+
+   * - short_wchar
+     - * 0 --- sizeof(wchar_t) == 4
+       * 1 --- sizeof(wchar_t) == 2
+
+   * - short_enum
+     - * 0 --- Enums are at least as large as an ``int``.
+       * 1 --- Enums are stored in the smallest integer type which can
+         represent all of its values.
+
+For example, the following metadata section specifies that the module was
+compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
+enum is the smallest type which can represent all of its values::
+
+    !llvm.module.flags = !{!0, !1}
+    !0 = metadata !{i32 1, metadata !"short_wchar", i32 1}
+    !1 = metadata !{i32 1, metadata !"short_enum", i32 0}
+
 .. _intrinsicglobalvariables:
 
 Intrinsic Global Variables
 ==========================
 
 LLVM has a number of "magic" global variables that contain data that
 affect code generation or other IR semantics. These are documented here.
 All globals of this sort should have a section specified as

include/llvm/Support/ARMBuildAttributes.h

enum {
   // Tag_ABI_PCS_RO_data, (=14), uleb128
   AddressROPCRel = 1, // Address RO static data PC-relative
   AddressRONone = 2, // No RO static data permitted
 
   // Tag_ABI_PCS_GOT_use, (=17), uleb128
   AddressDirect = 1, // Address imported data directly
   AddressGOT = 2, // Address imported data indirectly (via GOT)
 
+  // Tag_ABI_PCS_wchar_t, (=18), uleb128
+  WCharProhibited = 0,  // wchar_t is not used
+  WCharWidth2Bytes = 2, // sizeof(wchar_t) == 2
+  WCharWidth4Bytes = 4, // sizeof(wchar_t) == 4
+
   // Tag_ABI_FP_denormal, (=20), uleb128
   PreserveFPSign = 2, // sign when flushed-to-zero is preserved
 
   // Tag_ABI_FP_number_model, (=23), uleb128
   AllowRTABI = 2,  // numbers, infinities, and one quiet NaN (see [RTABI])
   AllowIEE754 = 3, // this code to use all the IEEE 754-defined FP encodings
 
+  // Tag_ABI_enum_size, (=26), uleb128
+  EnumProhibited = 0, // The user prohibited the use of enums when building
+                      // this entity.
+  EnumSmallest = 1,   // Enum is smallest container big enough to hold all
+                      // values.
+  Enum32Bit = 2,      // Enum is at least 32 bits.
+  Enum32BitABI = 3,   // Every enumeration visible across an ABI-complying
+                      // interface contains a value needing 32 bits to encode
+                      // it; other enums can be containerized.
+
   // Tag_ABI_HardFP_use, (=27), uleb128
   HardFPImplied = 0,          // FP use should be implied by Tag_FP_arch
   HardFPSinglePrecision = 1,  // Single-precision only
 
   // Tag_ABI_VFP_args, (=28), uleb128
   BaseAAPCS = 0,
   HardFPAAPCS = 1,
 

lib/Target/ARM/ARMAsmPrinter.cpp

void ARMAsmPrinter::emitAttributes() {
   // If only Thumb hwdiv is present, it must also be in base arch (ARMv7-R/M).
   // It is not possible to produce DisallowDIV: if hwdiv is present in the base
   // arch, supplying -hwdiv downgrades the effective arch, via ClearImpliedBits.
   // AllowDIVExt is only emitted if hwdiv isn't available in the base arch;
   // otherwise, the default value (AllowDIVIfExists) applies.
   if (Subtarget->hasDivideInARMMode() && !Subtarget->hasV8Ops())
       ATS.emitAttribute(ARMBuildAttrs::DIV_use, ARMBuildAttrs::AllowDIVExt);
 
+  if (MMI) {
+    if (const Module *SourceModule = MMI->getModule()) {
+      // ABI_PCS_wchar_t to indicate wchar_t width
+      if (auto WCharWidthValue = cast_or_null<ConstantInt>(
+              SourceModule->getModuleFlag("wchar_size"))) {
+        int WCharWidth = WCharWidthValue->getZExtValue();
+        assert((WCharWidth == 2 || WCharWidth == 4) &&
+               "wchar_t width must be 2 or 4 bytes");
+        ATS.emitAttribute(ARMBuildAttrs::ABI_PCS_wchar_t, WCharWidth);
+      }
+
+      // ABI_enum_size to indicate enum width
+      if (auto EnumWidthValue = cast_or_null<ConstantInt>(
+              SourceModule->getModuleFlag("min_enum_size"))) {
+        int EnumWidth = EnumWidthValue->getZExtValue();
+        assert((EnumWidth == 1 || EnumWidth == 4) &&
+               "Minimum enum width must be 1 or 4 bytes");
+        int EnumBuildAttr = EnumWidth == 1 ? 1 : 2;
+        ATS.emitAttribute(ARMBuildAttrs::ABI_enum_size, EnumBuildAttr);
+      }
+    }
+  }
+
   if (Subtarget->hasTrustZone() && Subtarget->hasVirtualization())
       ATS.emitAttribute(ARMBuildAttrs::Virtualization_use,
                         ARMBuildAttrs::AllowTZVirtualization);
   else if (Subtarget->hasTrustZone())
       ATS.emitAttribute(ARMBuildAttrs::Virtualization_use,
                         ARMBuildAttrs::AllowTZ);
   else if (Subtarget->hasVirtualization())
       ATS.emitAttribute(ARMBuildAttrs::Virtualization_use,

test/CodeGen/ARM/metadata-default.ll

+; RUN: llc < %s -mtriple=armv7-linux-gnueabi | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-n32-S64"
+target triple = "armv7--none-eabi"
+
+define i32 @f(i64 %z) {
+  ret i32 0
+}
+
+!llvm.module.flags = !{!0, !1}
+
+!0 = metadata !{i32 1, metadata !"wchar_size", i32 4}
+!1 = metadata !{i32 1, metadata !"min_enum_size", i32 4}
+
+; CHECK: .eabi_attribute 18, 4   @ Tag_ABI_PCS_wchar_t
+; CHECK: .eabi_attribute 26, 2   @ Tag_ABI_enum_size

test/CodeGen/ARM/metadata-short-enums.ll

+; RUN: llc < %s | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-n32-S64"
+target triple = "armv7--none-eabi"
+
+define i32 @f(i64 %z) {
+  ret i32 0
+}
+
+!llvm.module.flags = !{!0, !1}
+
+!0 = metadata !{i32 1, metadata !"wchar_size", i32 4}
+!1 = metadata !{i32 1, metadata !"min_enum_size", i32 1}
+
+; CHECK: .eabi_attribute 18, 4   @ Tag_ABI_PCS_wchar_t
+; CHECK: .eabi_attribute 26, 1   @ Tag_ABI_enum_size

test/CodeGen/ARM/metadata-short-wchar.ll

+; RUN: llc < %s | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-n32-S64"
+target triple = "armv7--none-eabi"
+
+define i32 @f(i64 %z) {
+  ret i32 0
+}
+
+!llvm.module.flags = !{!0, !1}
+
+!0 = metadata !{i32 1, metadata !"wchar_size", i32 2}
+!1 = metadata !{i32 1, metadata !"min_enum_size", i32 4}
+
+; CHECK: .eabi_attribute 18, 2   @ Tag_ABI_PCS_wchar_t
+; CHECK: .eabi_attribute 26, 2   @ Tag_ABI_enum_size