target/arm: Implement MVE VMULL (polynomial)

Implement the MVE VMULL (polynomial) insn.  Unlike Neon, this comes
in two flavours: 8x8->16 and a 16x16->32.  Also unlike Neon, the
inputs are in either the low or the high half of each double-width
element.

The assembler for this insn indicates the size with "P8" or "P16",
encoded into bit 28 as size = 0 or 1. We choose to follow the
same encoding as VQDMULL and decode this into a->size as MO_16
or MO_32 indicating the size of the result elements. This then
carries through to the helper function names where it then
matches up with the existing pmull_h() which does an 8x8->16
operation and a new pmull_w() which does the 16x16->32.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>

6 files changed, 83 insertions, 5 deletions

diff --git a/target/arm/helper-mve.h b/target/arm/helper-mve.h
index 56e4084..84adfb2 100644
--- a/target/arm/helper-mve.h
+++ b/target/arm/helper-mve.h
@@ -145,6 +145,11 @@ DEF_HELPER_FLAGS_4(mve_vmulltub, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
 DEF_HELPER_FLAGS_4(mve_vmulltuh, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
 DEF_HELPER_FLAGS_4(mve_vmulltuw, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
 
+DEF_HELPER_FLAGS_4(mve_vmullpbh, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
+DEF_HELPER_FLAGS_4(mve_vmullpth, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
+DEF_HELPER_FLAGS_4(mve_vmullpbw, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
+DEF_HELPER_FLAGS_4(mve_vmullptw, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
+
 DEF_HELPER_FLAGS_4(mve_vqdmulhb, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
 DEF_HELPER_FLAGS_4(mve_vqdmulhh, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
 DEF_HELPER_FLAGS_4(mve_vqdmulhw, TCG_CALL_NO_WG, void, env, ptr, ptr, ptr)
diff --git a/target/arm/mve.decode b/target/arm/mve.decode
index fa9d921..de079ec 100644
--- a/target/arm/mve.decode
+++ b/target/arm/mve.decode
@@ -173,10 +173,16 @@ VHADD_U          111 1 1111 0 . .. ... 0 ... 0 0000 . 1 . 0 ... 0 @2op
 VHSUB_S          111 0 1111 0 . .. ... 0 ... 0 0010 . 1 . 0 ... 0 @2op
 VHSUB_U          111 1 1111 0 . .. ... 0 ... 0 0010 . 1 . 0 ... 0 @2op
 
-VMULL_BS         111 0 1110 0 . .. ... 1 ... 0 1110 . 0 . 0 ... 0 @2op
-VMULL_BU         111 1 1110 0 . .. ... 1 ... 0 1110 . 0 . 0 ... 0 @2op
-VMULL_TS         111 0 1110 0 . .. ... 1 ... 1 1110 . 0 . 0 ... 0 @2op
-VMULL_TU         111 1 1110 0 . .. ... 1 ... 1 1110 . 0 . 0 ... 0 @2op
+{
+  VMULLP_B       111 . 1110 0 . 11 ... 1 ... 0 1110 . 0 . 0 ... 0 @2op_sz28
+  VMULL_BS       111 0 1110 0 . .. ... 1 ... 0 1110 . 0 . 0 ... 0 @2op
+  VMULL_BU       111 1 1110 0 . .. ... 1 ... 0 1110 . 0 . 0 ... 0 @2op
+}
+{
+  VMULLP_T       111 . 1110 0 . 11 ... 1 ... 1 1110 . 0 . 0 ... 0 @2op_sz28
+  VMULL_TS       111 0 1110 0 . .. ... 1 ... 1 1110 . 0 . 0 ... 0 @2op
+  VMULL_TU       111 1 1110 0 . .. ... 1 ... 1 1110 . 0 . 0 ... 0 @2op
+}
 
 VQDMULH          1110 1111 0 . .. ... 0 ... 0 1011 . 1 . 0 ... 0 @2op
 VQRDMULH         1111 1111 0 . .. ... 0 ... 0 1011 . 1 . 0 ... 0 @2op
diff --git a/target/arm/mve_helper.c b/target/arm/mve_helper.c
index be8b954..91fb346 100644
--- a/target/arm/mve_helper.c
+++ b/target/arm/mve_helper.c
@@ -482,6 +482,22 @@ DO_2OP_L(vmulltuh, 1, 2, uint16_t, 4, uint32_t, DO_MUL)
 DO_2OP_L(vmulltuw, 1, 4, uint32_t, 8, uint64_t, DO_MUL)
 
 /*
+ * Polynomial multiply. We can always do this generating 64 bits
+ * of the result at a time, so we don't need to use DO_2OP_L.
+ */
+#define VMULLPH_MASK 0x00ff00ff00ff00ffULL
+#define VMULLPW_MASK 0x0000ffff0000ffffULL
+#define DO_VMULLPBH(N, M) pmull_h((N) & VMULLPH_MASK, (M) & VMULLPH_MASK)
+#define DO_VMULLPTH(N, M) DO_VMULLPBH((N) >> 8, (M) >> 8)
+#define DO_VMULLPBW(N, M) pmull_w((N) & VMULLPW_MASK, (M) & VMULLPW_MASK)
+#define DO_VMULLPTW(N, M) DO_VMULLPBW((N) >> 16, (M) >> 16)
+
+DO_2OP(vmullpbh, 8, uint64_t, DO_VMULLPBH)
+DO_2OP(vmullpth, 8, uint64_t, DO_VMULLPTH)
+DO_2OP(vmullpbw, 8, uint64_t, DO_VMULLPBW)
+DO_2OP(vmullptw, 8, uint64_t, DO_VMULLPTW)
+
+/*
  * Because the computation type is at least twice as large as required,
  * these work for both signed and unsigned source types.
  */
diff --git a/target/arm/translate-mve.c b/target/arm/translate-mve.c
index a2a4503..d318f34 100644
--- a/target/arm/translate-mve.c
+++ b/target/arm/translate-mve.c
@@ -464,6 +464,34 @@ static bool trans_VQDMULLT(DisasContext *s, arg_2op *a)
     return do_2op(s, a, fns[a->size]);
 }
 
+static bool trans_VMULLP_B(DisasContext *s, arg_2op *a)
+{
+    /*
+     * Note that a->size indicates the output size, ie VMULL.P8
+     * is the 8x8->16 operation and a->size is MO_16; VMULL.P16
+     * is the 16x16->32 operation and a->size is MO_32.
+     */
+    static MVEGenTwoOpFn * const fns[] = {
+        NULL,
+        gen_helper_mve_vmullpbh,
+        gen_helper_mve_vmullpbw,
+        NULL,
+    };
+    return do_2op(s, a, fns[a->size]);
+}
+
+static bool trans_VMULLP_T(DisasContext *s, arg_2op *a)
+{
+    /* a->size is as for trans_VMULLP_B */
+    static MVEGenTwoOpFn * const fns[] = {
+        NULL,
+        gen_helper_mve_vmullpth,
+        gen_helper_mve_vmullptw,
+        NULL,
+    };
+    return do_2op(s, a, fns[a->size]);
+}
+
 /*
  * VADC and VSBC: these perform an add-with-carry or subtract-with-carry
  * of the 32-bit elements in each lane of the input vectors, where the
diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c
index 034f6b8..17fb158 100644
--- a/target/arm/vec_helper.c
+++ b/target/arm/vec_helper.c
@@ -2028,11 +2028,23 @@ static uint64_t expand_byte_to_half(uint64_t x)
          | ((x & 0xff000000) << 24);
 }
 
-static uint64_t pmull_h(uint64_t op1, uint64_t op2)
+uint64_t pmull_w(uint64_t op1, uint64_t op2)
 {
     uint64_t result = 0;
     int i;
+    for (i = 0; i < 16; ++i) {
+        uint64_t mask = (op1 & 0x0000000100000001ull) * 0xffffffff;
+        result ^= op2 & mask;
+        op1 >>= 1;
+        op2 <<= 1;
+    }
+    return result;
+}
 
+uint64_t pmull_h(uint64_t op1, uint64_t op2)
+{
+    uint64_t result = 0;
+    int i;
     for (i = 0; i < 8; ++i) {
         uint64_t mask = (op1 & 0x0001000100010001ull) * 0xffff;
         result ^= op2 & mask;
diff --git a/target/arm/vec_internal.h b/target/arm/vec_internal.h
index 865d213..2a33558 100644
--- a/target/arm/vec_internal.h
+++ b/target/arm/vec_internal.h
@@ -206,4 +206,15 @@ int16_t do_sqrdmlah_h(int16_t, int16_t, int16_t, bool, bool, uint32_t *);
 int32_t do_sqrdmlah_s(int32_t, int32_t, int32_t, bool, bool, uint32_t *);
 int64_t do_sqrdmlah_d(int64_t, int64_t, int64_t, bool, bool);
 
+/*
+ * 8 x 8 -> 16 vector polynomial multiply where the inputs are
+ * in the low 8 bits of each 16-bit element
+*/
+uint64_t pmull_h(uint64_t op1, uint64_t op2);
+/*
+ * 16 x 16 -> 32 vector polynomial multiply where the inputs are
+ * in the low 16 bits of each 32-bit element
+ */
+uint64_t pmull_w(uint64_t op1, uint64_t op2);
+
 #endif /* TARGET_ARM_VEC_INTERNALS_H */