We dont actually need vzeroupper at start of function

GCC emits vzeroupper automatically when compiling with native
optimizations but we still need it otherwise

kitty/simd-string-impl.h

@@ -297,7 +297,6 @@ FUNC(zero_last_n_bytes)(integer_t vec, const integer_t index, char n) {
 }
 
 #define find_match(haystack, sz, get_test_vec) { \
-    /* zero_upper(); this is disabled as some buggy compilers re-order instructions causing the test result vec to be zero after the test */\
     const uint8_t* limit = haystack + sz; \
     integer_t chunk; int n; \
 \

tools/simdstring/generate.go

@@ -1153,10 +1153,6 @@ func (s *Function) OutputASM(w io.Writer) {
 	s.print_signature(w)
 	fmt.Fprintf(w, "\nTEXT ·%s(SB), NOSPLIT, $0-%d\n", s.Name, s.Size)
 
-	if s.Used256BitReg {
-		fmt.Fprintln(w, "\tVZEROUPPER // zero upper bits of AVX registers to avoid dependencies when switching between SSE and AVX code")
-	}
-
 	has_trailing_return := false
 	for _, i := range s.Instructions {
 		if len(i) == 0 {