… mf2 length.

Merge back from release branch to copy 0.3.32 tag

This reverts commit 7eab365 and fixes
the pattern rule requirement "% must match a non-empty stem" to match at
least '.' in the pattern.

This whole file could actually be substantially simplified to just:

	%.$(SUFFIX): %.c
		$(CC) $(CFLAGS) -c $< -o $(@f)

	%.$(PSUFFIX): %.c
		$(CC) $(PFLAGS) -c $< -o $(@f)

if desired to entirely avoid the copy-paste duplication, but the net
effect is the same.

Fixes: 42c7a27

…ations

c_check: loongarch64: Fix typo

This change fixes a regression in SBGEMM where C is assumed to be BF16,
and so unconditionally casts the output to FP32 resulting in incorrect
outputs when beta=1.

Fix incorrect cast from BF16 to FP32 in SBGEMM

typedef the unsupported fp16 as bfloat16 on Loongarch64 too

[ARM64] Add optimized fp16 shgemm kernels for Neoverse N2

Automatically assume BIGNUMA if more than 256 CPUs defined/detected

Add CPU ID for Neoverse V3

Add WebAssembly SIMD STRSM and DTRSM kernels

Add part code for NeoverseV3 to the build-time detection as well

Guard against eventual overflow of the config string

Disable gcc-specific version checks for NVIDIA HPC in CMake builds

Optimize ?LACPY/?LASCL/?LANTR looping in fat matrix cases with UPLO=L (Reference-LAPACK PR 1251)

…r SME

Fix ARM64 DYNAMIC_ARCH unconditionally branching into the SME code for STRMM

Undefine GEMM_PREFERRED_SIZE before redefining to silence warning

Fix CMake builds on Apple that use the Ninja generator

Fix EXTERNAL declarations in CLARF1F and ZUPMTR  (Reference-LAPACK PR 1257)

Update the Changelog for 0.3.33