erasure_code: fix vmovdqa instruction

vmovdqa needs to be vmovdqa32/64 when used on ZMMs (EVEX encoded).

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>

erasure_code/gf_2vect_dot_prod_avx512.asm

@@ -133,8 +133,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_2vect_mad_avx512.asm

@@ -118,8 +118,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_3vect_dot_prod_avx512.asm

@@ -142,8 +142,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_3vect_mad_avx512.asm

@@ -117,8 +117,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_4vect_dot_prod_avx512.asm

@@ -156,8 +156,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_4vect_mad_avx512.asm

@@ -117,8 +117,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_5vect_dot_prod_avx512.asm

@@ -172,8 +172,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_5vect_mad_avx512.asm

@@ -120,8 +120,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_6vect_dot_prod_avx512.asm

@@ -172,8 +172,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_6vect_mad_avx512.asm

@@ -129,8 +129,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_vect_dot_prod_avx512.asm

@@ -104,8 +104,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq

erasure_code/gf_vect_mad_avx512.asm

@@ -101,8 +101,8 @@
 %else
 ;;; Use Non-temporal load/stor
  %ifdef NO_NT_LDST
-  %define XLDR vmovdqa
-  %define XSTR vmovdqa
+  %define XLDR vmovdqa64
+  %define XSTR vmovdqa64
  %else
   %define XLDR vmovntdqa
   %define XSTR vmovntdq