Megahit in Arm Neoverse de 64 bits core

[higaredavm]

We are exploring bioinformtics in amazon AWS, specifically we want to run genome assembly on AWS Graviton2 processors which have an ARM architecture. Is there a version of megahit built for ARM / aarch64 architecture.

[aclum]

We have had good success with graviton2 with other bioinformatics tools and would also be interested of a version of megathit which compiles on ARM.

[higaredavm]

We have had good success with graviton2 with other bioinformatics tools and would also be interested of a version of megathit which compiles on ARM.

I am curious about what bioinformatics tools have you been running on graviton2.

[aclum]

From BBTools, BBMap and BBcms both run on graviton2. These are AWS blog posts on peregrine and bwa on graviton2
https://aws.amazon.com/blogs/publicsector/accelerating-genome-assembly-aws-graviton2/
https://aws.amazon.com/blogs/publicsector/generalized-approach-benchmarking-genomics-workloads-cloud-bwa-read-aligner-graviton2/

[aclum]

I made some progress on getting MEGAHIT working on ARM but am stalled on the third issue

The first issue is with x86 specific flags (-mbmi2 & -mpopcnt) which can be removed from the CMakeLists.txt file.

The second issue is with x86intrin.h and can be fixed by replacing with openvec*.h files from https://github.com/OpenVec/OpenVec

The third issue is an error about "impossible constraint in ‘asm’". Based on searching stack overflow this can be fixed by rewriting the code in c++ instead of assembly
https://stackoverflow.com/questions/1478513/linux-assembler-error-impossible-constraint-in-asm

Would it be possible to get the third issue fixed?

[voutcn]

You can just replace

megahit/src/utils/cpu_dispatch.h

Lines 8 to 34 in 7dde1ca

	inline bool HasPopcnt() {
	unsigned eax, ebx, ecx, edx;
	#ifdef _MSC_VER
	int cpuid[4];
	__cpuid(cpuid, 1);
	eax = cpuid[0], ebx = cpuid[1], ecx = cpuid[2], edx = cpuid[3];
	#else
	asm volatile("cpuid\n\t"
	: "=a"(eax), "=b"(ebx), "=c"(ecx), "=d"(edx)
	: "0"(1));
	#endif
	return ecx >> 23U & 1U;
	}
	inline bool HasBmi2() {
	unsigned eax, ebx, ecx, edx;
	#ifdef _MSC_VER
	int cpuid[4];
	__cpuidex(cpuid, 7, 0);
	eax = cpuid[0], ebx = cpuid[1], ecx = cpuid[2], edx = cpuid[3];
	#else
	asm volatile("cpuid\n\t"
	: "=a"(eax), "=b"(ebx), "=c"(ecx), "=d"(edx)
	: "0"(7), "2"(0));
	#endif
	return ebx >> 8U & 1U;
	}

with

inline bool HasPopcnt() { return false; }
inline bool HasBmi2() { return false; }

I don't have an Arm machine to test. Let me know if that works. Also feel free to send a PR.

@aquaskyline to follow up.

[accopeland]

For the record, the above suggestions are sufficient to compile megahit on graviton2. I am working on a PR.

[higaredavm]

@voutcn We are doing the changes that you suggest, however we are having this error. I really appreciate your help, thanks

[voutcn]

I think the error has nothing to do with ARM. One *.edges.info file is broken for some unknown reason. Could you retry?

[martin-g]

Hi!

I created a new PR - #368
It is built on top of #329 but preserves the x86_64 optimizations.