Multiple ways to change gemmPrecision

[jerinphilip]

--gem-precision and --int* are two ways to do the same thing. Functionality would still work and be accessible without the following.

marian-dev/src/common/config_parser.cpp

Lines 933 to 947 in 844800e

	cli.add<bool>("--int8",
	"Optimize speed even more aggressively sacrificing memory or precision by using 8bit integer GEMM with intgemm instead of floats. Only available on CPU. Corresponds to --gemm-precision int8");
	cli.add<bool>("--int8Alpha",
	"Use a precomputed quantisation multipliers for the activations. Requires a special model. Corresponds to --gemm-precision int8Alpha");
	cli.add<bool>("--int8shift",
	"Use a faster, shifted integer 8bit GEMM implementation. Corresponds to --gemm-precision int8shift");
	cli.add<bool>("--int8shiftAlpha",
	"Use a faster, shifted integer 8bit GEMM implementation, with precomputed alphas. Corresponds to --gemm-precision int8shiftAlpha");
	cli.add<bool>("--int8shiftAll",
	"Use a faster, shifted integer 8bit GEMM implementation even for matrices that don't have a bias. Beneficial on VNNI. Corresponds to --gemm-precision int8shiftAll");
	cli.add<bool>("--int8shiftAlphaAll",
	"Use a faster, shifted integer 8bit GEMM implementation even for matrices that don't have a bias, with precomputed alphas. Should be the fastest option. Corresponds to --gemm-precision int8shiftAlphaAll");
	cli.add<std::string>("--gemm-precision",
	"Use lower precision for the GEMM operations only. Supported values: float32, int16, int8, int8Alpha, int8shift, int8shiftAlpha, int8shiftAll, int8shiftAlphaAll", "float32");
	cli.add<bool>("--dump-quantmult",