Updated SAIL for tile loads/stores when (rs2) == 0

Author: joseemoreira

src/integrated-matrix.adoc

@@ -2638,6 +2638,7 @@ Encoding::
 Description::
 Loads a 2D matrix tile from memory into the vector register group starting at _vd_.
 _rs1_ holds the matrix base address; _rs2_ holds the leading dimension `LD` in elements.
+if _rs2_ holds zero (0), the leading dimension `LD` is set to λ × LMUL.
 When _Lλ_ is specified, it overrides `lambda[2:0]` from `vtype` for this instruction.
 Inactive elements (masked off) are not updated unless `vtype.vma`=1 (mask-agnostic), in which case they may be overwritten with 1s.
 Tail elements (index ≥ VL) are not updated unless `vtype.vta`=1 (tail-agnostic), in which case they may be overwritten with 1s.
@@ -2677,6 +2678,10 @@ let LD     : int      = unsigned(X(rs2));
 let vm_val            = read_vmask(num_elem, vm, zvreg);
 let linesize : int    = eff_lambda * LMUL;
 
+if LD == 0 then {
+  LD = eff_lambda * LMUL;
+}
+
 foreach (i from 0 to (num_elem - 1)) {
   if vm_val[i] == 0b1 then {
     set_vstart(to_bits_unsafe(16, i));
@@ -2737,6 +2742,7 @@ Encoding::
 Description::
 Stores the 2D matrix tile held in the vector register group starting at _vs3_ to memory.
 _rs1_ holds the matrix base address; _rs2_ holds the leading dimension `LD` in elements.
+if _rs2_ holds zero (0), the leading dimension `LD` is set to λ × LMUL.
 When _Lλ_ is specified, it overrides `lambda[2:0]` from `vtype` for this instruction.
 Inactive elements (masked off) are not written to memory and do not raise exceptions.
 Tail elements (index ≥ VL) are not written to memory and do not raise exceptions.
@@ -2770,6 +2776,10 @@ let LD     : int   = unsigned(X(rs2));
 let vm_val         = read_vmask(num_elem, vm, zvreg);
 let linesize : int = eff_lambda * LMUL;
 
+if LD == 0 then {
+  LD = eff_lambda * LMUL;
+}
+
 foreach (i from 0 to (num_elem - 1)) {
   if vm_val[i] == 0b1 then {
     set_vstart(to_bits_unsafe(16, i));
@@ -2833,6 +2843,7 @@ required by the matrix multiplier.
 This instruction is used when a B tile is stored in row-major order, or when an A or C tile
 is stored in column-major order in memory.
 _rs1_ holds the matrix base address; _rs2_ holds the leading dimension `LD` in elements.
+if _rs2_ holds zero (0), the leading dimension `LD` is set to λ × LMUL.
 When _Lλ_ is specified, it overrides `lambda[2:0]` from `vtype` for this instruction.
 Inactive elements (masked off) are not updated unless `vtype.vma`=1 (mask-agnostic), in which case they may be overwritten with 1s.
 Tail elements (index ≥ VL) are not updated unless `vtype.vta`=1 (tail-agnostic), in which case they may be overwritten with 1s.
@@ -2867,6 +2878,10 @@ let LD     : int   = unsigned(X(rs2));
 let vm_val         = read_vmask(num_elem, vm, zvreg);
 let linesize : int = eff_lambda * LMUL;
 
+if LD == 0 then {
+  LD = eff_lambda * LMUL;
+}
+
 foreach (i from 0 to (num_elem - 1)) {
   if vm_val[i] == 0b1 then {
     set_vstart(to_bits_unsafe(16, i));
@@ -2929,6 +2944,7 @@ Description::
 Stores a 2D matrix tile from vector registers to memory, applying the inverse transposition
 of `vmttl.v`.
 _rs1_ holds the matrix base address; _rs2_ holds the leading dimension `LD` in elements.
+if _rs2_ holds zero (0), the leading dimension `LD` is set to λ × LMUL.
 When _Lλ_ is specified, it overrides `lambda[2:0]` from `vtype` for this instruction.
 Inactive elements (masked off) are not written to memory and do not raise exceptions.
 Tail elements (index ≥ VL) are not written to memory and do not raise exceptions.
@@ -2962,6 +2978,10 @@ let LD     : int   = unsigned(X(rs2));
 let vm_val         = read_vmask(num_elem, vm, zvreg);
 let linesize : int = eff_lambda * LMUL;
 
+if LD == 0 then {
+  LD = eff_lambda * LMUL;
+}
+
 foreach (i from 0 to (num_elem - 1)) {
   if vm_val[i] == 0b1 then {
     set_vstart(to_bits_unsafe(16, i));