Use ".arch_extension" ARM directive to support hwdiv on krait

[sgundapa]

Summary

In case of "krait" CPU, asm printer doesn't emit any ".cpu" so the
features bits are not computed. This patch lets the asm printer
emit ".cpu cortex-a9" directive for krait and the hwdiv feature is
enabled through ".arch_extension". In short, krait is treated
as "cortex-a9" with hwdiv. We can not emit ".krait" as CPU since
it is not supported bu GNU GAS yet.

[sgundapa]

The .arch_extension implementation patch is here: http://reviews.llvm.org/D7316

[rengolin]

lib/Target/ARM/ARMAsmPrinter.cpp

709

I was looking around and emitting plain text of a hardcoded value here would be a bad idea.

I think you need to create/append an enum and do like FPU/CPU/ARCH.

[sgundapa]

lib/Target/ARM/ARMAsmPrinter.cpp

709

Renato, Can you be more specifc on this ?
If you can point me to an existing implementation, it would be nice.

[rengolin]

lib/Target/ARM/ARMAsmPrinter.cpp

709

see emitFPU().

Valid values for name are the same as those accepted as architectural extensions by the -mcpu commandline option.

The same duplication that emitFPU has introduced (by adding the ARMFPUName.def) will need to be introduced, and that will all go away once we unify the parsers. For now, let's have it in the same way, so I can wipe them out all at once later.

[sgundapa]

Hopefully, this addresses everyone's concerns

[rengolin]

With the comment inline sorted, LGTM, thanks!

lib/Target/ARM/ARMAsmPrinter.cpp

707

Can you move this up with the first check? I think it'll be easier to "remove Krait" later, as the FIXME hints to.

[sgundapa]

lib/Target/ARM/ARMAsmPrinter.cpp

707

Sure. I will do that and commit this patch.
Thanks for your patience.

[echristo]

Sure.

-eric

[sgundapa]

This patch is breaking the mips bot here.
Any idea why ?
http://lab.llvm.org:8011/builders/llvm-mips-linux/builds/12343/steps/configure/logs/stdio

[sgundapa]

Thanks for the fix Renato

lib/Target/ARM/ARMAsmPrinter.cpp

 
   std::string CPUString = STI.getCPUString();
 
-  // FIXME: remove krait check when GNU tools support krait cpu
-  if (CPUString != "generic" && CPUString != "krait")
-    ATS.emitTextAttribute(ARMBuildAttrs::CPU_name, CPUString);
+  if (CPUString != "generic") {
+    // FIXME: remove krait check when GNU tools support krait cpu
+    if (STI.isKrait())
+      ATS.emitTextAttribute(ARMBuildAttrs::CPU_name, "cortex-a9");
+    else
+      ATS.emitTextAttribute(ARMBuildAttrs::CPU_name, CPUString);
+  }
 
   ATS.emitAttribute(ARMBuildAttrs::CPU_arch, getArchForCPU(CPUString, &STI));
 

                       ARMBuildAttrs::AllowThumb32);
   }
 
+  // We consider krait as a "cortex-a9" + hwdiv CPU
+  // Enable hwdiv through ".arch_extension idiv"
+  if (STI.isKrait() && (STI.hasDivide() || STI.hasDivideInARMMode()))
+    ATS.emitArchExtension("idiv");
+
   if (STI.hasNEON()) {
     /* NEON is not exactly a VFP architecture, but GAS emit one of
      * neon/neon-fp-armv8/neon-vfpv4/vfpv3/vfpv2 for .fpu parameters */

test/CodeGen/ARM/krait-cpu-div-attribute.ll

+; Tests the genration of ".arch_extension" attribute for hardware
+; division on krait CPU. For now, krait is recognized as "cortex-a9" + hwdiv
+; Also, tests for the hwdiv instruction on krait CPU
+
+; check for arch_extension/cpu directive
+; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=krait | FileCheck %s --check-prefix=DIV_EXTENSION
+; RUN: llc < %s -mtriple=thumbv7-linux-gnueabi -mcpu=krait | FileCheck %s --check-prefix=DIV_EXTENSION
+; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a9 | FileCheck %s --check-prefix=NODIV_KRAIT
+; RUN: llc < %s -mtriple=thumbv7-linux-gnueabi -mcpu=cortex-a9 | FileCheck %s --check-prefix=NODIV_KRAIT
+; RUN: llc < %s -mcpu=krait -mattr=-hwdiv,-hwdiv-arm | FileCheck %s --check-prefix=NODIV_KRAIT
+
+; check if correct instruction is emitted by integrated assembler
+; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=krait -filetype=obj | llvm-objdump -mcpu=krait -triple armv7-linux-gnueabi -d - | FileCheck %s --check-prefix=HWDIV
+; RUN: llc < %s -mtriple=thumbv7-linux-gnueabi -mcpu=krait -filetype=obj | llvm-objdump -mcpu=krait -triple thumbv7-linux-gnueabi -d - | FileCheck %s --check-prefix=HWDIV
+
+; arch_extension attribute
+; DIV_EXTENSION:  .cpu cortex-a9
+; DIV_EXTENSION:  .arch_extension idiv
+; NODIV_KRAIT-NOT:  .arch_extension idiv
+; HWDIV: sdiv
+
+define i32 @main() #0 {
+entry:
+  %retval = alloca i32, align 4
+  %a = alloca i32, align 4
+  %b = alloca i32, align 4
+  %c = alloca i32, align 4
+  store i32 0, i32* %retval
+  store volatile i32 100, i32* %b, align 4
+  store volatile i32 32, i32* %c, align 4
+  %0 = load volatile i32* %b, align 4
+  %1 = load volatile i32* %c, align 4
+  %div = sdiv i32 %0, %1
+  store volatile i32 %div, i32* %a, align 4
+  ret i32 0
+}