[Qemu-devel] [PATCH 40/40] Add vsumsws, vsum2sws, and vsum4{sbs, shs, ubs} instructions.

[Nathan Froyd]

Signed-off-by: Nathan Froyd <address@hidden>
---
 target-ppc/helper.h    |    5 ++
 target-ppc/op_helper.c |  103 ++++++++++++++++++++++++++++++++++++++++++++++++
 target-ppc/translate.c |    5 ++
 3 files changed, 113 insertions(+), 0 deletions(-)

diff --git a/target-ppc/helper.h b/target-ppc/helper.h
index 1dd2cf8..51760a1 100644
--- a/target-ppc/helper.h
+++ b/target-ppc/helper.h
@@ -217,6 +217,11 @@ DEF_HELPER_2(lvewx, void, avr, tl)
 DEF_HELPER_2(stvebx, void, avr, tl)
 DEF_HELPER_2(stvehx, void, avr, tl)
 DEF_HELPER_2(stvewx, void, avr, tl)
+DEF_HELPER_3(vsumsws, void, avr, avr, avr)
+DEF_HELPER_3(vsum2sws, void, avr, avr, avr)
+DEF_HELPER_3(vsum4sbs, void, avr, avr, avr)
+DEF_HELPER_3(vsum4shs, void, avr, avr, avr)
+DEF_HELPER_3(vsum4ubs, void, avr, avr, avr)
 
 DEF_HELPER_1(efscfsi, i32, i32)
 DEF_HELPER_1(efscfui, i32, i32)
diff --git a/target-ppc/op_helper.c b/target-ppc/op_helper.c
index 5c45072..d72b011 100644
--- a/target-ppc/op_helper.c
+++ b/target-ppc/op_helper.c
@@ -2667,6 +2667,109 @@ void helper_vsubcuw (ppc_avr_t *r, ppc_avr_t *a, 
ppc_avr_t *b)
     }
 }
 
+void helper_vsumsws (ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
+{
+    int64_t t;
+    int i, upper;
+    ppc_avr_t result;
+    int sat = 0;
+
+#if defined(WORDS_BIGENDIAN)
+    upper = ARRAY_SIZE(r->s32)-1;
+#else
+    upper = 0;
+#endif
+    t = (int64_t)b->s32[upper];
+    for (i = 0; i < ARRAY_SIZE(r->s32); i++) {
+        t += a->s32[i];
+        result.s32[i] = 0;
+    }
+    result.s32[upper] = cvtsdsw(t, &sat);
+    *r = result;
+
+    if (sat) {
+        env->vscr |= (1 << VSCR_SAT);
+    }
+}
+
+void helper_vsum2sws (ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
+{
+    int i, j, upper;
+    ppc_avr_t result;
+    int sat = 0;
+
+#if defined(WORDS_BIGENDIAN)
+    upper = 1;
+#else
+    upper = 0;
+#endif
+    for (i = 0; i < ARRAY_SIZE(r->u64); i++) {
+        int64_t t = (int64_t)b->s32[upper+i*2];
+        result.u64[i] = 0;
+        for (j = 0; j < ARRAY_SIZE(r->u64); j++) {
+            t += a->s32[2*i+j];
+        }
+        result.s32[upper+i*2] = cvtsdsw(t, &sat);
+    }
+
+    *r = result;
+    if (sat) {
+        env->vscr |= (1 << VSCR_SAT);
+    }
+}
+
+void helper_vsum4sbs (ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
+{
+    int i, j;
+    int sat = 0;
+
+    for (i = 0; i < ARRAY_SIZE(r->s32); i++) {
+        int64_t t = (int64_t)b->s32[i];
+        for (j = 0; j < ARRAY_SIZE(r->s32); j++) {
+            t += a->s8[4*i+j];
+        }
+        r->s32[i] = cvtsdsw(t, &sat);
+    }
+
+    if (sat) {
+        env->vscr |= (1 << VSCR_SAT);
+    }
+}
+
+void helper_vsum4shs (ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
+{
+    int sat = 0;
+    int i;
+
+    for (i = 0; i < ARRAY_SIZE(r->s32); i++) {
+        int64_t t = (int64_t)b->s32[i];
+        t += a->s16[2*i] + a->s16[2*i+1];
+        r->s32[i] = cvtsdsw(t, &sat);
+    }
+
+    if (sat) {
+        env->vscr |= (1 << VSCR_SAT);
+    }
+}
+
+void helper_vsum4ubs (ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)
+{
+    int i, j;
+    int sat = 0;
+
+    for (i = 0; i < ARRAY_SIZE(r->u32); i++) {
+        uint64_t t = (uint64_t)b->u32[i];
+        for (j = 0; j < ARRAY_SIZE(r->u32); j++) {
+            t += a->u8[4*i+j];
+        }
+        r->u32[i] = cvtuduw(t, &sat);
+    }
+
+    if (sat) {
+        env->vscr |= (1 << VSCR_SAT);
+    }
+}
+
 #if defined(WORDS_BIGENDIAN)
 #define UPKHI 1
 #define UPKLO 0
diff --git a/target-ppc/translate.c b/target-ppc/translate.c
index 2ec9d19..593c5dd 100644
--- a/target-ppc/translate.c
+++ b/target-ppc/translate.c
@@ -6377,6 +6377,11 @@ GEN_VXFORM(vpkswus, 7, 5);
 GEN_VXFORM(vpkshss, 7, 6);
 GEN_VXFORM(vpkswss, 7, 7);
 GEN_VXFORM(vpkpx, 7, 12);
+GEN_VXFORM(vsum4ubs, 4, 24);
+GEN_VXFORM(vsum4sbs, 4, 28);
+GEN_VXFORM(vsum4shs, 4, 25);
+GEN_VXFORM(vsum2sws, 4, 26);
+GEN_VXFORM(vsumsws, 4, 30);
 
 #define GEN_VXFORM_NOA(name, opc2, opc3)                                \
     GEN_HANDLER(name, 0x04, opc2, opc3, 0x001f0000, PPC_ALTIVEC)        \
-- 
1.6.0.5