arch-power: Add fixed-point doubleword multiply instructions

This adds the following arithmetic instructions:
  * Multiply Low Doubleword (mulld[o][.])
  * Multiply High Doubleword (mulhd[.])
  * Multiply High Doubleword Unsigned (mulhdu[.])

Change-Id: I505d94dc8e9711c575c94f75e10f7e05e1d05fdf
Signed-off-by: Sandipan Das <sandipan@linux.vnet.ibm.com>

diff --git a/src/arch/power/insts/integer.hh b/src/arch/power/insts/integer.hh
index 4ffd4691f3ccd0225b1decad70ec078f2b2a500f..6a34184a28a5a4865216d8e83afea2c22c977c1e 100644 (file)
--- a/src/arch/power/insts/integer.hh
+++ b/src/arch/power/insts/integer.hh
@@ -156,6 +156,46 @@ class IntArithOp : public IntOp
     {
     }
 
+    /* Compute 128-bit product of 64-bit unsigned integer multiplication */
+    inline std::tuple<uint64_t, uint64_t>
+    multiply(uint64_t ra, uint64_t rb) const
+    {
+        uint64_t plo, phi;
+    #if defined(__SIZEOF_INT128__)
+        __uint128_t prod = (__uint128_t)ra * rb;
+        plo = prod;
+        phi = prod >> 64;
+    #else
+        uint64_t ralo = (uint32_t)ra, rahi = ra >> 32;
+        uint64_t rblo = (uint32_t)rb, rbhi = rb >> 32;
+        uint64_t pp0 = ralo * rblo;
+        uint64_t pp1 = rahi * rblo;
+        uint64_t pp2 = ralo * rbhi;
+        uint64_t pp3 = rahi * rbhi;
+        uint64_t c = ((uint32_t)pp1) + ((uint32_t)pp2) + (pp0 >> 32);
+        phi = pp3 + (pp2 >> 32) + (pp1 >> 32) + (c >> 32);
+        plo = (c << 32) | ((uint32_t)pp0);
+    #endif
+        return std::make_tuple(plo, phi);
+    }
+
+    /* Compute 128-bit product of 64-bit signed integer multiplication */
+    inline std::tuple<uint64_t, int64_t>
+    multiply(int64_t ra, int64_t rb) const
+    {
+        uint64_t plo, phi;
+    #if defined(__SIZEOF_INT128__)
+        __int128_t prod = (__int128_t)ra * rb;
+        plo = prod;
+        phi = prod >> 64;
+    #else
+        std::tie(plo, phi) = multiply((uint64_t)ra, (uint64_t)rb);
+        if (rb < 0) phi -= (uint64_t)ra;
+        if (ra < 0) phi -= (uint64_t)rb;
+    #endif
+        return std::make_tuple(plo, (int64_t)phi);
+    }
+
     std::string generateDisassembly(
             Addr pc, const SymbolTable *symtab) const override;
 };

diff --git a/src/arch/power/isa/decoder.isa b/src/arch/power/isa/decoder.isa
index cc90f8537b78c077642fcd5fb32738184904f36c..f7ae7ffb280ee1c83565d82e0b169c36bc794c6b 100644 (file)
--- a/src/arch/power/isa/decoder.isa
+++ b/src/arch/power/isa/decoder.isa
@@ -611,6 +611,30 @@ decode PO default Unknown::unknown() {
                 }},
                 true);
 
+                73: mulhd({{
+                    int64_t res;
+                    std::tie(std::ignore, res) = multiply(Ra_sd, Rb_sd);
+                    Rt = res;
+                }});
+
+                9: mulhdu({{
+                    uint64_t res;
+                    std::tie(std::ignore, res) = multiply(Ra, Rb);
+                    Rt = res;
+                }});
+
+                233: mulld({{
+                    int64_t src1 = Ra_sd;
+                    int64_t src2 = Rb_sd;
+                    uint64_t res = src1 * src2;
+                    std::tie(res, std::ignore) = multiply(src1, src2);
+                    if (src1 != 0 && (int64_t)res / src1 != src2) {
+                        setOV = true;
+                    }
+                    Rt = res;
+                }},
+                true);
+
                 491: divw({{
                     int32_t src1 = Ra_sw;
                     int32_t src2 = Rb_sw;