Re: [PATCH v4 23/45] target/arm: Implement SME ADDHA, ADDVA

[Richard Henderson]

On 7/4/22 16:20, Peter Maydell wrote:

+void HELPER(sme_addha_d)(void *vzda, void *vzn, void *vpn,
+                         void *vpm, uint32_t desc)
+{
+    intptr_t row, col, oprsz = simd_oprsz(desc) / 8;
+    uint8_t *pn = vpn, *pm = vpm;
+    uint64_t *zda = vzda, *zn = vzn;
+
+    for (row = 0; row < oprsz; ++row) {
+        if (pn[H1(row)] & 1) {
+            for (col = 0; col < oprsz; ++col) {
+                if (pm[H1(col)] & 1) {
+                    zda[row * sizeof(ARMVectorReg) + col] += zn[col];
+                }
+            }
+        }
+    }
+}

These array index calculations look wrong again?
Should be 'row * (sizeof(ARMVectorReg) / 8) + col' or equivalent,
I think.

The tile rows are N physical rows apart.


r~