Extend snrt_dma_memset

[colluca]

The snrt_memset function in alloc.h:

snitch_cluster/sw/snRuntime/src/alloc.h

Lines 98 to 103 in 1b9da15

	// TODO colluca: optimize by using DMA
	inline void *snrt_memset(void *ptr, int value, size_t num) {
	for (uint32_t i = 0; i < num; ++i)
	*((uint8_t *)ptr + i) = (unsigned char)value;
	return ptr;
	}

And the snrt_dma_memset function in dma.h are to some extent redundant:

snitch_cluster/sw/snRuntime/src/dma.h

Lines 389 to 408 in 1b9da15

	/**
	* @brief Fast memset function performed by DMA.
	* @param ptr Pointer to the start of the region.
	* @param value Value to set.
	* @param len Number of bytes, must be a multiple of the DMA bus width.
	*/
	inline void snrt_dma_memset(void *ptr, uint8_t value, uint32_t len) {
	// set first 64bytes to value
	// memset(ptr, value, 64);
	uint8_t *p = ptr;
	uint32_t nbytes = 64;
	while (nbytes--) {
	*p++ = value;
	}
	// DMA copy the the rest
	snrt_dma_txid_t memset_txid =
	snrt_dma_start_2d(ptr, ptr, 64, 64, 0, len / 64);
	snrt_dma_wait_all();
	}

Merge them and also handle specific case of a memset to zero by taking advantage of the cluster's zero memory. See the following as an example:

snitch_cluster/sw/snRuntime/src/start.c

Lines 40 to 41 in 1b9da15

	snrt_dma_start_1d((void*)(tls_ptr + i * tls_offset),
	(void*)(snrt_zero_memory_ptr()), size);