Added 1 more opcode

ptitSeb · Oct 3, 2024 · 5ebdd07 · 5ebdd07
1 parent c16d512
commit 5ebdd07
Showing 1 changed file with 29 additions and 13 deletions.
diff --git a/src/dynarec/rv64/dynarec_rv64_660f_vector.c b/src/dynarec/rv64/dynarec_rv64_660f_vector.c
@@ -1767,27 +1767,43 @@ uintptr_t dynarec64_660F_vector(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t i
             VSLL_VX(q0, q0, x4, VECTOR_UNMASKED);
             break;
         case 0xF5:
-            if (rv64_xtheadvector) { DEFAULT_VECTOR; } // lack of vrgatherei16.vv
-
             INST_NAME("PMADDWD Gx, Ex");
             nextop = F8;
             SET_ELEMENT_WIDTH(x1, VECTOR_SEW16, 1);
             GETGX_vector(q0, 1, VECTOR_SEW16);
             GETEX_vector(q1, 0, 0, VECTOR_SEW16);
             v0 = fpu_get_scratch_lmul(dyn, VECTOR_LMUL2);
             v1 = fpu_get_scratch_lmul(dyn, VECTOR_LMUL2);
-            VWMUL_VV(v0, q1, q0, VECTOR_UNMASKED);
+            d1 = fpu_get_scratch(dyn);                    // use this at caution!
             d0 = fpu_get_scratch_lmul(dyn, VECTOR_LMUL2); // warning, no more scratches!
-            ADDI(x4, xZR, 6);
-            VID_V(d0, VECTOR_UNMASKED);
-            VSLL_VI(d0, d0, 1, VECTOR_UNMASKED); // times 2
-            VMIN_VX(d0, d0, x4, VECTOR_UNMASKED);
-            VADD_VI(q0, d0, 1, VECTOR_UNMASKED);
-            vector_vsetvli(dyn, ninst, x1, VECTOR_SEW32, VECTOR_LMUL2, 2);
-            VRGATHEREI16_VV(v1, v0, d0, VECTOR_UNMASKED); // 6 4 2 0
-            VRGATHEREI16_VV(d0, v0, q0, VECTOR_UNMASKED); // 7 5 3 1
-            SET_ELEMENT_WIDTH(x1, VECTOR_SEW32, 1);
-            VADD_VV(q0, v1, d0, VECTOR_UNMASKED);
+            VWMUL_VV(v0, q1, q0, VECTOR_UNMASKED);
+            if (rv64_xtheadvector) { // lack of vrgatherei16.vv
+                ADDI(x4, xZR, 6);
+                vector_vsetvli(dyn, ninst, x1, VECTOR_SEW32, VECTOR_LMUL2, 1);
+                VID_V(d0, VECTOR_UNMASKED);
+                VSLL_VI(d0, d0, 1, VECTOR_UNMASKED); // times 2
+                VMIN_VX(d0, d0, x4, VECTOR_UNMASKED);
+                VRGATHER_VV(v1, v0, d0, VECTOR_UNMASKED); // 6 4 2 0
+                // out of scratches, go back to lmul1 and slideup....
+                vector_vsetvli(dyn, ninst, x1, VECTOR_SEW32, VECTOR_LMUL1, 1);
+                VSLIDEUP_VX(d1, v1, 0, VECTOR_UNMASKED);
+                vector_vsetvli(dyn, ninst, x1, VECTOR_SEW32, VECTOR_LMUL2, 1);
+                VADD_VI(d0, d0, 1, VECTOR_UNMASKED);
+                VRGATHER_VV(v1, v0, d0, VECTOR_UNMASKED); // 7 5 3 1
+                SET_ELEMENT_WIDTH(x1, VECTOR_SEW32, 1);
+                VADD_VV(q0, v1, d1, VECTOR_UNMASKED);
+            } else {
+                ADDI(x4, xZR, 6);
+                VID_V(d0, VECTOR_UNMASKED);
+                VSLL_VI(d0, d0, 1, VECTOR_UNMASKED); // times 2
+                VMIN_VX(d0, d0, x4, VECTOR_UNMASKED);
+                VADD_VI(q0, d0, 1, VECTOR_UNMASKED);
+                vector_vsetvli(dyn, ninst, x1, VECTOR_SEW32, VECTOR_LMUL2, 1);
+                VRGATHEREI16_VV(v1, v0, d0, VECTOR_UNMASKED); // 6 4 2 0
+                VRGATHEREI16_VV(d0, v0, q0, VECTOR_UNMASKED); // 7 5 3 1
+                SET_ELEMENT_WIDTH(x1, VECTOR_SEW32, 1);
+                VADD_VV(q0, v1, d0, VECTOR_UNMASKED);
+            }
             break;
         case 0xF6:
             INST_NAME("PSADBW Gx, Ex");