target-tilegx: Implement complex multiply instructions

Signed-off-by: Richard Henderson <rth@twiddle.net>

3 files changed, 73 insertions, 1 deletions

diff --git a/target-tilegx/helper.c b/target-tilegx/helper.c
index cad5daeaae..36b287f84a 100644
--- a/target-tilegx/helper.c
+++ b/target-tilegx/helper.c
@@ -97,3 +97,43 @@ uint64_t helper_crc32_32(uint64_t accum, uint64_t input)
     /* zlib crc32 converts the accumulator and output to one's complement.  */
     return crc32(accum ^ 0xffffffff, buf, 4) ^ 0xffffffff;
 }
+
+uint64_t helper_cmula(uint64_t srcd, uint64_t srca, uint64_t srcb)
+{
+    uint32_t reala = (int16_t)srca;
+    uint32_t imaga = (int16_t)(srca >> 16);
+    uint32_t realb = (int16_t)srcb;
+    uint32_t imagb = (int16_t)(srcb >> 16);
+    uint32_t reald = srcd;
+    uint32_t imagd = srcd >> 32;
+    uint32_t realr = reala * realb - imaga * imagb + reald;
+    uint32_t imagr = reala * imagb + imaga * realb + imagd;
+
+    return deposit64(realr, 32, 32, imagr);
+}
+
+uint64_t helper_cmulaf(uint64_t srcd, uint64_t srca, uint64_t srcb)
+{
+    uint32_t reala = (int16_t)srca;
+    uint32_t imaga = (int16_t)(srca >> 16);
+    uint32_t realb = (int16_t)srcb;
+    uint32_t imagb = (int16_t)(srcb >> 16);
+    uint32_t reald = (int16_t)srcd;
+    uint32_t imagd = (int16_t)(srcd >> 16);
+    int32_t realr = reala * realb - imaga * imagb;
+    int32_t imagr = reala * imagb + imaga * realb;
+
+    return deposit32((realr >> 15) + reald, 16, 16, (imagr >> 15) + imagd);
+}
+
+uint64_t helper_cmul2(uint64_t srca, uint64_t srcb, int shift, int round)
+{
+    uint32_t reala = (int16_t)srca;
+    uint32_t imaga = (int16_t)(srca >> 16);
+    uint32_t realb = (int16_t)srcb;
+    uint32_t imagb = (int16_t)(srcb >> 16);
+    int32_t realr = reala * realb - imaga * imagb + round;
+    int32_t imagr = reala * imagb + imaga * realb + round;
+
+    return deposit32(realr >> shift, 16, 16, imagr >> shift);
+}
diff --git a/target-tilegx/helper.h b/target-tilegx/helper.h
index 72c8e9297c..82d84f1ae7 100644
--- a/target-tilegx/helper.h
+++ b/target-tilegx/helper.h
@@ -6,6 +6,9 @@ DEF_HELPER_FLAGS_1(revbits, TCG_CALL_NO_RWG_SE, i64, i64)
 DEF_HELPER_FLAGS_3(shufflebytes, TCG_CALL_NO_RWG_SE, i64, i64, i64, i64)
 DEF_HELPER_FLAGS_2(crc32_8, TCG_CALL_NO_RWG_SE, i64, i64, i64)
 DEF_HELPER_FLAGS_2(crc32_32, TCG_CALL_NO_RWG_SE, i64, i64, i64)
+DEF_HELPER_FLAGS_3(cmula, TCG_CALL_NO_RWG_SE, i64, i64, i64, i64)
+DEF_HELPER_FLAGS_3(cmulaf, TCG_CALL_NO_RWG_SE, i64, i64, i64, i64)
+DEF_HELPER_FLAGS_4(cmul2, TCG_CALL_NO_RWG_SE, i64, i64, i64, int, int)
 
 DEF_HELPER_FLAGS_2(v1multu, TCG_CALL_NO_RWG_SE, i64, i64, i64)
 DEF_HELPER_FLAGS_2(v1shl, TCG_CALL_NO_RWG_SE, i64, i64, i64)
diff --git a/target-tilegx/translate.c b/target-tilegx/translate.c
index 9f86fd3710..d7e4d526e2 100644
--- a/target-tilegx/translate.c
+++ b/target-tilegx/translate.c
@@ -276,6 +276,15 @@ static void gen_mul_half(TCGv tdest, TCGv tsrca, TCGv tsrcb,
     tcg_temp_free(t);
 }
 
+static void gen_cmul2(TCGv tdest, TCGv tsrca, TCGv tsrcb, int sh, int rd)
+{
+    TCGv_i32 tsh = tcg_const_i32(sh);
+    TCGv_i32 trd = tcg_const_i32(rd);
+    gen_helper_cmul2(tdest, tsrca, tsrcb, tsh, trd);
+    tcg_temp_free_i32(tsh);
+    tcg_temp_free_i32(trd);
+}
+
 static TileExcp gen_st_opcode(DisasContext *dc, unsigned dest, unsigned srca,
                               unsigned srcb, TCGMemOp memop, const char *name)
 {
@@ -759,13 +768,33 @@ static TileExcp gen_rrr_opcode(DisasContext *dc, unsigned opext,
         mnemonic = "cmpne";
         break;
     case OE_RRR(CMULAF, 0, X0):
+        gen_helper_cmulaf(tdest, load_gr(dc, dest), tsrca, tsrcb);
+        mnemonic = "cmulaf";
+        break;
     case OE_RRR(CMULA, 0, X0):
+        gen_helper_cmula(tdest, load_gr(dc, dest), tsrca, tsrcb);
+        mnemonic = "cmula";
+        break;
     case OE_RRR(CMULFR, 0, X0):
+        gen_cmul2(tdest, tsrca, tsrcb, 15, 1 << 14);
+        mnemonic = "cmulfr";
+        break;
     case OE_RRR(CMULF, 0, X0):
+        gen_cmul2(tdest, tsrca, tsrcb, 15, 0);
+        mnemonic = "cmulf";
+        break;
     case OE_RRR(CMULHR, 0, X0):
+        gen_cmul2(tdest, tsrca, tsrcb, 16, 1 << 15);
+        mnemonic = "cmulhr";
+        break;
     case OE_RRR(CMULH, 0, X0):
+        gen_cmul2(tdest, tsrca, tsrcb, 16, 0);
+        mnemonic = "cmulh";
+        break;
     case OE_RRR(CMUL, 0, X0):
-        return TILEGX_EXCP_OPCODE_UNIMPLEMENTED;
+        gen_helper_cmula(tdest, load_zero(dc), tsrca, tsrcb);
+        mnemonic = "cmul";
+        break;
     case OE_RRR(CRC32_32, 0, X0):
         gen_helper_crc32_32(tdest, tsrca, tsrcb);
         mnemonic = "crc32_32";