ggml : simplify the ggml_compute_forward_ calls

[ggerganov]

There is no need to explicitly list all the source tensors - they can be accessed through dst when needed:

ggml/src/ggml.c

Lines 14783 to 14812 in 6b14d73

	switch (tensor->op) {
	case GGML_OP_DUP:
	{
	ggml_compute_forward_dup(params, tensor->src[0], tensor);
	} break;
	case GGML_OP_ADD:
	{
	ggml_compute_forward_add(params, tensor->src[0], tensor->src[1], tensor);
	} break;
	case GGML_OP_ADD1:
	{
	ggml_compute_forward_add1(params, tensor->src[0], tensor->src[1], tensor);
	} break;
	case GGML_OP_ACC:
	{
	ggml_compute_forward_acc(params, tensor->src[0], tensor->src[1], tensor);
	} break;
	case GGML_OP_SUB:
	{
	ggml_compute_forward_sub(params, tensor->src[0], tensor->src[1], tensor);
	} break;
	case GGML_OP_MUL:
	{
	ggml_compute_forward_mul(params, tensor->src[0], tensor->src[1], tensor);
	} break;
	case GGML_OP_DIV:
	{
	ggml_compute_forward_div(params, tensor->src[0], tensor->src[1], tensor);
	} break;
	case GGML_OP_SQR:

[siddharthvader]

PR #729 addresses this and should be ready now

[ggerganov]

Thanks!