Avoid use of __int128_t

It is nonstandard, and GCC doesn't support it on 32-bit platforms. The
resulting code for MULH[[S]U] is crappier, but that doesn't really matter,
as these instructions are dynamically infrequent.

11 files changed, 54 insertions, 22 deletions

diff --git a/hwacha/insn_template_hwacha_ut.h b/hwacha/insn_template_hwacha_ut.h
index a26e528..d585f9b 100644
--- a/hwacha/insn_template_hwacha_ut.h
+++ b/hwacha/insn_template_hwacha_ut.h
@@ -1,4 +1,5 @@
 #include "hwacha.h"
+#include "mulhi.h"
 #include "decode_hwacha_ut.h"
 #include "softfloat.h"
 #include "platform.h" // softfloat isNaNF32UI, etc.
diff --git a/riscv/cachesim.h b/riscv/cachesim.h
index 51044c3..a529007 100644
--- a/riscv/cachesim.h
+++ b/riscv/cachesim.h
@@ -7,7 +7,7 @@
 #include <cstring>
 #include <string>
 #include <map>
-#include <stdint.h>
+#include <cstdint>
 
 class lfsr_t
 {
diff --git a/riscv/decode.h b/riscv/decode.h
index d647b2c..b32c6e4 100644
--- a/riscv/decode.h
+++ b/riscv/decode.h
@@ -7,17 +7,13 @@
 # error spike requires a two''s-complement c++ implementation
 #endif
 
-#define __STDC_LIMIT_MACROS
-#include <stdint.h>
+#include <cstdint>
 #include <string.h>
 #include "encoding.h"
 #include "config.h"
 #include "common.h"
 #include <cinttypes>
 
-typedef int int128_t __attribute__((mode(TI)));
-typedef unsigned int uint128_t __attribute__((mode(TI)));
-
 typedef int64_t sreg_t;
 typedef uint64_t reg_t;
 typedef uint64_t freg_t;
diff --git a/riscv/dummy-rocc-test.c b/riscv/dummy-rocc-test.c
index ba362a4..4071ece 100644
--- a/riscv/dummy-rocc-test.c
+++ b/riscv/dummy-rocc-test.c
@@ -5,7 +5,7 @@
 
 #include <assert.h>
 #include <stdio.h>
-#include <stdint.h>
+#include <cstdint>
 
 int main() {
   uint64_t x = 123, y = 456, z = 0;
diff --git a/riscv/insn_template.h b/riscv/insn_template.h
index 109b4c4..56a9c7f 100644
--- a/riscv/insn_template.h
+++ b/riscv/insn_template.h
@@ -1,4 +1,5 @@
 #include "mmu.h"
+#include "mulhi.h"
 #include "softfloat.h"
 #include "platform.h" // softfloat isNaNF32UI, etc.
 #include "internals.h" // ditto
diff --git a/riscv/insns/mulh.h b/riscv/insns/mulh.h
index f63869d..8ae7520 100644
--- a/riscv/insns/mulh.h
+++ b/riscv/insns/mulh.h
@@ -1,8 +1,4 @@
-if(xpr64)
-{
-  int64_t a = RS1;
-  int64_t b = RS2;
-  WRITE_RD((int128_t(a) * int128_t(b)) >> 64);
-}
+if (xpr64)
+  WRITE_RD(mulh(RS1, RS2));
 else
   WRITE_RD(sext32((sext32(RS1) * sext32(RS2)) >> 32));
diff --git a/riscv/insns/mulhsu.h b/riscv/insns/mulhsu.h
index d62256e..3168ade 100644
--- a/riscv/insns/mulhsu.h
+++ b/riscv/insns/mulhsu.h
@@ -1,8 +1,4 @@
-if(xpr64)
-{
-  int64_t a = RS1;
-  uint64_t b = RS2;
-  WRITE_RD((int128_t(a) * uint128_t(b)) >> 64);
-}
+if (xpr64)
+  WRITE_RD(mulhsu(RS1, RS2));
 else
   WRITE_RD(sext32((sext32(RS1) * reg_t((uint32_t)RS2)) >> 32));
diff --git a/riscv/insns/mulhu.h b/riscv/insns/mulhu.h
index 2d6f48c..b03b870 100644
--- a/riscv/insns/mulhu.h
+++ b/riscv/insns/mulhu.h
@@ -1,4 +1,4 @@
-if(xpr64)
-  WRITE_RD((uint128_t(RS1) * uint128_t(RS2)) >> 64);
+if (xpr64)
+  WRITE_RD(mulhu(RS1, RS2));
 else
   WRITE_RD(sext32(((uint64_t)(uint32_t)RS1 * (uint64_t)(uint32_t)RS2) >> 32));
diff --git a/riscv/memtracer.h b/riscv/memtracer.h
index e223c43..f2f1b1c 100644
--- a/riscv/memtracer.h
+++ b/riscv/memtracer.h
@@ -3,7 +3,7 @@
 #ifndef _MEMTRACER_H
 #define _MEMTRACER_H
 
-#include <stdint.h>
+#include <cstdint>
 #include <string.h>
 #include <vector>
 
diff --git a/riscv/mulhi.h b/riscv/mulhi.h
new file mode 100644
index 0000000..f0a5d7b
--- /dev/null
+++ b/riscv/mulhi.h
@@ -0,0 +1,41 @@
+#ifndef _RISCV_MULHI_H
+#define _RISCV_MULHI_H
+
+#include <cstdint>
+
+inline uint64_t mulhu(uint64_t a, uint64_t b)
+{
+  uint64_t t;
+  uint32_t y1, y2, y3;
+  uint64_t a0 = (uint32_t)a, a1 = a >> 32;
+  uint64_t b0 = (uint32_t)b, b1 = b >> 32;
+
+  t = a1*b0 + ((a0*b0) >> 32);
+  y1 = t;
+  y2 = t >> 32;
+
+  t = a0*b1 + y1;
+  y1 = t;
+
+  t = a1*b1 + y2 + (t >> 32);
+  y2 = t;
+  y3 = t >> 32;
+
+  return ((uint64_t)y3 << 32) | y2;
+}
+
+inline int64_t mulh(int64_t a, int64_t b)
+{
+  int negate = (a < 0) != (b < 0);
+  uint64_t res = mulhu(a < 0 ? -a : a, b < 0 ? -b : b);
+  return negate ? ~res + (a * b == 0) : res;
+}
+
+inline int64_t mulhsu(int64_t a, uint64_t b)
+{
+  int negate = a < 0;
+  uint64_t res = mulhu(a < 0 ? -a : a, b);
+  return negate ? ~res + (a * b == 0) : res;
+}
+
+#endif
diff --git a/riscv/riscv.mk.in b/riscv/riscv.mk.in
index 0fd5eaf..e9ca63f 100644
--- a/riscv/riscv.mk.in
+++ b/riscv/riscv.mk.in
@@ -21,6 +21,7 @@ riscv_hdrs = \
 	rocc.h \
 	dummy-rocc.h \
 	insn_template.h \
+	mulhi.h \
 
 riscv_precompiled_hdrs = \
 	insn_template.h \