[arm] Add softvfp to supported FPU names.

[logan]

Summary

In this patch, I am going to add "softvfp" to supported ARM FPU names.

Note: The "softvfp" is a FPU target which will not generate any build
attributes.

[logan]

Fix the unittest.

[rengolin]

Hi Logan,

Shouldn't it set the VFP build attributes to disabled?

[logan]

According to the "Addenda to the ABI for the ARM architecture", the default value for Tag_FP_arch is not allowed, which means that the generated object should not use VFP instructions. Thus, I think it is OK to omit this build attribute.

BTW, I am not very sure for the situation for Tag_DIV_use. It seems that Tag_DIV_use is not related to softvfp. The latest GNU assembler does not emit this build attribute, but the older version did so. Should I add this attribute for softvfp? Thanks.

[rengolin]

Richard can answer that question better than me.

[logan]

Ping.

[richard.barton.arm]

Hello Logan, Renato

Sorry for the late reply, I have been away for holidays.

• The TAG_FP_arch attribute should be omitted or set to 0 for softvfp, so the patch is correct in that respect. The difference between softfp and hardfp should be in the Tag_ABI_VFP_args attribute. If the other directives for .fpu imply hard floating point, then they should define this attribute with value 1. I don't know if gcc does this already, but it should.
• Tag_DIV_use is indeed orthogonal to soft/hard float and should not be affected by these directives.

Assuming that this makes llvm compatible with gcc vis Tag_ABI_VFP_args then I see nothing wrong.

[logan]

Thanks for your review. Committed as 198313.

lib/Target/ARM/ARMFPUName.def

 ARM_FPU_NAME("vfpv3-d16", VFPV3_D16)
 ARM_FPU_NAME("vfpv4", VFPV4)
 ARM_FPU_NAME("vfpv4-d16", VFPV4_D16)
 ARM_FPU_NAME("fp-armv8", FP_ARMV8)
 ARM_FPU_NAME("neon", NEON)
 ARM_FPU_NAME("neon-vfpv4", NEON_VFPV4)
 ARM_FPU_NAME("neon-fp-armv8", NEON_FP_ARMV8)
 ARM_FPU_NAME("crypto-neon-fp-armv8", CRYPTO_NEON_FP_ARMV8)
+ARM_FPU_NAME("softvfp", SOFTVFP)
 
 #undef ARM_FPU_NAME

lib/Target/ARM/MCTargetDesc/ARMELFStreamer.cpp

case ARM::CRYPTO_NEON_FP_ARMV8:
     setAttributeItem(ARMBuildAttrs::VFP_arch,
                      ARMBuildAttrs::AllowFPARMv8A,
                      /* OverwriteExisting= */ false);
     setAttributeItem(ARMBuildAttrs::Advanced_SIMD_arch,
                      ARMBuildAttrs::AllowNeonARMv8,
                      /* OverwriteExisting= */ false);
     break;
 
+  case ARM::SOFTVFP:
+    break;
+
   default:
     report_fatal_error("Unknown FPU: " + Twine(FPU));
     break;
   }
 }
 size_t ARMTargetELFStreamer::calculateContentSize() const {
   size_t Result = 0;
   for (size_t i = 0; i < Contents.size(); ++i) {

test/MC/ARM/directive-fpu-softvfp.s

+@ RUN: llvm-mc < %s -triple armv7-unknown-linux-gnueabi \
+@ RUN:              -filetype=obj -o /dev/null 2>&1
+
+  .fpu  softvfp