3 files changed, 5 insertions, 10 deletions
diff --git a/src/dynarec/arm64/dynarec_arm64_0f.c b/src/dynarec/arm64/dynarec_arm64_0f.c
index 5d983399..7850393d 100644
--- a/src/dynarec/arm64/dynarec_arm64_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_0f.c
@@ -2814,8 +2814,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             GETEM(v1, 0);
             q0 = fpu_get_scratch(dyn, ninst);
             VUMULL_16(q0, v0, v1);
-            VSHRQ_32(q0, q0, 16);
-            XTN_16(v0, q0);
+            UQSHRN_16(v0, q0, 16);  // saturation will never happens as only 16bits remain and fits in 16bits
             break;
         case 0xE5:
             INST_NAME("PMULHW Gm,Em");
diff --git a/src/dynarec/arm64/dynarec_arm64_660f.c b/src/dynarec/arm64/dynarec_arm64_660f.c
index d36dea7b..49477f16 100644
--- a/src/dynarec/arm64/dynarec_arm64_660f.c
+++ b/src/dynarec/arm64/dynarec_arm64_660f.c
@@ -3108,10 +3108,8 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             q1 = fpu_get_scratch(dyn, ninst);
             VUMULL_16(q0, v0, v1);
             VUMULL2_16(q1, v0, v1);
-            VSHRQ_32(q0, q0, 16);
-            VSHRQ_32(q1, q1, 16);
-            XTN_16(v0, q0);
-            XTN2_16(v0, q1);
+            UQSHRN_16(v0, q0, 16);  // 16bits->16bits: no saturation
+            UQSHRN2_16(v0, q1, 16);
             break;
         case 0xE5:
             INST_NAME("PMULHW Gx, Ex");
diff --git a/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c b/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c
index aea2f378..e878056d 100644
--- a/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_avx_66_0f.c
@@ -1626,10 +1626,8 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                 if(!l) { GETGX_empty_VXEX(v0, v2, v1, 0); } else { GETGY_empty_VYEY(v0, v2, v1); }
                 VUMULL_16(q0, v2, v1);
                 VUMULL2_16(q1, v2, v1);
-                VSHRQ_32(q0, q0, 16);
-                VSHRQ_32(q1, q1, 16);
-                XTN_16(v0, q0);
-                XTN2_16(v0, q1);
+                UQSHRN_16(v0, q0, 16);  // 16bits->16bits: no saturation
+                UQSHRN2_16(v0, q1, 16);
             }
             if(!vex.l) YMM0(gd);
             break;