260212.173057.CET revise norma/fortran/linalg.f90 to include the change to the scaling in norm

Author: zaikunzhang

.development

@@ -2546,3 +2546,10 @@ xcrun
 dsrc
 ncall
 ccc
+AMOP
+ATf
+emax
+emin
+scalmax
+scalmin
+sumx

.github/actions/spelling/allow.txt

@@ -39,7 +39,7 @@ module linalg_mod
 !
 ! Started: July 2020
 !
-! Last Modified: Thu 12 Feb 2026 04:02:55 PM CET
+! Last Modified: Thu 12 Feb 2026 05:30:33 PM CET
 !--------------------------------------------------------------------------------------------------!
 
 implicit none
@@ -2032,6 +2032,8 @@ function named_norm_mat(x, nname) result(y)
 ! Calculation starts !
 !====================!
 
+! N.B.: Ideally, we should also do a scaling similar to that in P_NORM to avoid over/underflow.
+
 if (size(x, 1) * size(x, 2) == 0) then
     y = ZERO
 elseif (.not. all(is_finite(x))) then

fortran/common/linalg.f90

@@ -23,9 +23,10 @@ if [[ -f "$LINALG" ]] ; then
     STR="Y is NaN if and only if X contains NaN"
     $SEDI "/$STR/d" "$LINALG"
 
-    # See https://github.com/llvm/llvm-project/issues/180957
-    STR="Y >= 0 unless X contains NaN"
-    $SEDI "/$STR/d" "$LINALG"
+    # The following should have been fixed by commit 399df3ae055566c3f6133118ebab6fdcc16cebb5
+    # # See https://github.com/llvm/llvm-project/issues/180957
+    # STR="Y >= 0 unless X contains NaN"
+    # $SEDI "/$STR/d" "$LINALG"
 
     # # With flang 20.1.8, matprod in linalg.f90 often fails to compute matrix multiplications
     # # correctly, generating garbage values (NaN or numbers that are nearly zero). No idea why.