Merge remote-tracking branch 'remotes/rth-gitlab/tags/pull-hex-20210502' into staging

Minor cleanups.
Finish the rest of the hexagon integer instructions.

# gpg: Signature made Sun 02 May 2021 15:38:17 BST
# gpg:                using RSA key 7A481E78868B4DB6A85A05C064DF38E8AF7E215F
# gpg:                issuer "richard.henderson@linaro.org"
# gpg: Good signature from "Richard Henderson <richard.henderson@linaro.org>" [full]
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth-gitlab/tags/pull-hex-20210502: (31 commits)
  Hexagon (target/hexagon) CABAC decode bin
  Hexagon (target/hexagon) load into shifted register instructions
  Hexagon (target/hexagon) load and unpack bytes instructions
  Hexagon (target/hexagon) bit reverse (brev) addressing
  Hexagon (target/hexagon) circular addressing
  Hexagon (target/hexagon) add A4_addp_c/A4_subp_c
  Hexagon (target/hexagon) add A6_vminub_RdP
  Hexagon (target/hexagon) add A5_ACS (vacsh)
  Hexagon (target/hexagon) add F2_sfinvsqrta
  Hexagon (target/hexagon) add F2_sfrecipa instruction
  Hexagon (target/hexagon) compile all debug code
  Hexagon (target/hexagon) move QEMU_GENERATE to only be on during macros.h
  Hexagon (target/hexagon) cleanup reg_field_info definition
  Hexagon (target/hexagon) cleanup ternary operators in semantics
  Hexagon (target/hexagon) use softfloat for float-to-int conversions
  Hexagon (target/hexagon) replace float32_mul_pow2 with float32_scalbn
  Hexagon (target/hexagon) use softfloat default NaN and tininess
  Hexagon (target/hexagon) change type of softfloat_roundingmodes
  Hexagon (target/hexagon) remove unused carry_from_add64 function
  Hexagon (target/hexagon) change variables from int to bool when appropriate
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

8 files changed, 2142 insertions, 0 deletions

diff --git a/tests/tcg/hexagon/Makefile.target b/tests/tcg/hexagon/Makefile.target
index 616af69..0992787 100644
--- a/tests/tcg/hexagon/Makefile.target
+++ b/tests/tcg/hexagon/Makefile.target
@@ -28,6 +28,7 @@ endif
 
 
 CFLAGS += -Wno-incompatible-pointer-types -Wno-undefined-internal
+CFLAGS += -fno-unroll-loops
 
 HEX_SRC=$(SRC_PATH)/tests/tcg/hexagon
 VPATH += $(HEX_SRC)
@@ -39,7 +40,12 @@ HEX_TESTS = first
 HEX_TESTS += misc
 HEX_TESTS += preg_alias
 HEX_TESTS += dual_stores
+HEX_TESTS += multi_result
 HEX_TESTS += mem_noshuf
+HEX_TESTS += circ
+HEX_TESTS += brev
+HEX_TESTS += load_unpack
+HEX_TESTS += load_align
 HEX_TESTS += atomics
 HEX_TESTS += fpstuff
 
diff --git a/tests/tcg/hexagon/brev.c b/tests/tcg/hexagon/brev.c
new file mode 100644
index 0000000..9736a24
--- /dev/null
+++ b/tests/tcg/hexagon/brev.c
@@ -0,0 +1,190 @@
+/*
+ *  Copyright(c) 2019-2021 Qualcomm Innovation Center, Inc. All Rights Reserved.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <stdio.h>
+#include <string.h>
+
+int err;
+
+#define NBITS          8
+#define SIZE           (1 << NBITS)
+
+long long     dbuf[SIZE] __attribute__((aligned(1 << 16))) = {0};
+int           wbuf[SIZE] __attribute__((aligned(1 << 16))) = {0};
+short         hbuf[SIZE] __attribute__((aligned(1 << 16))) = {0};
+unsigned char bbuf[SIZE] __attribute__((aligned(1 << 16))) = {0};
+
+/*
+ * We use the C preporcessor to deal with the combinations of types
+ */
+
+#define BREV_LOAD(SZ, RES, ADDR, INC) \
+    __asm__( \
+        "m0 = %2\n\t" \
+        "%0 = mem" #SZ "(%1++m0:brev)\n\t" \
+        : "=r"(RES), "+r"(ADDR) \
+        : "r"(INC) \
+        : "m0")
+
+#define BREV_LOAD_b(RES, ADDR, INC) \
+    BREV_LOAD(b, RES, ADDR, INC)
+#define BREV_LOAD_ub(RES, ADDR, INC) \
+    BREV_LOAD(ub, RES, ADDR, INC)
+#define BREV_LOAD_h(RES, ADDR, INC) \
+    BREV_LOAD(h, RES, ADDR, INC)
+#define BREV_LOAD_uh(RES, ADDR, INC) \
+    BREV_LOAD(uh, RES, ADDR, INC)
+#define BREV_LOAD_w(RES, ADDR, INC) \
+    BREV_LOAD(w, RES, ADDR, INC)
+#define BREV_LOAD_d(RES, ADDR, INC) \
+    BREV_LOAD(d, RES, ADDR, INC)
+
+#define BREV_STORE(SZ, PART, ADDR, VAL, INC) \
+    __asm__( \
+        "m0 = %2\n\t" \
+        "mem" #SZ "(%0++m0:brev) = %1" PART "\n\t" \
+        : "+r"(ADDR) \
+        : "r"(VAL), "r"(INC) \
+        : "m0", "memory")
+
+#define BREV_STORE_b(ADDR, VAL, INC) \
+    BREV_STORE(b, "", ADDR, VAL, INC)
+#define BREV_STORE_h(ADDR, VAL, INC) \
+    BREV_STORE(h, "", ADDR, VAL, INC)
+#define BREV_STORE_f(ADDR, VAL, INC) \
+    BREV_STORE(h, ".H", ADDR, VAL, INC)
+#define BREV_STORE_w(ADDR, VAL, INC) \
+    BREV_STORE(w, "", ADDR, VAL, INC)
+#define BREV_STORE_d(ADDR, VAL, INC) \
+    BREV_STORE(d, "", ADDR, VAL, INC)
+
+#define BREV_STORE_NEW(SZ, ADDR, VAL, INC) \
+    __asm__( \
+        "m0 = %2\n\t" \
+        "{\n\t" \
+        "    r5 = %1\n\t" \
+        "    mem" #SZ "(%0++m0:brev) = r5.new\n\t" \
+        "}\n\t" \
+        : "+r"(ADDR) \
+        : "r"(VAL), "r"(INC) \
+        : "r5", "m0", "memory")
+
+#define BREV_STORE_bnew(ADDR, VAL, INC) \
+    BREV_STORE_NEW(b, ADDR, VAL, INC)
+#define BREV_STORE_hnew(ADDR, VAL, INC) \
+    BREV_STORE_NEW(h, ADDR, VAL, INC)
+#define BREV_STORE_wnew(ADDR, VAL, INC) \
+    BREV_STORE_NEW(w, ADDR, VAL, INC)
+
+int bitreverse(int x)
+{
+    int result = 0;
+    int i;
+    for (i = 0; i < NBITS; i++) {
+        result <<= 1;
+        result |= x & 1;
+        x >>= 1;
+    }
+    return result;
+}
+
+int sext8(int x)
+{
+    return (x << 24) >> 24;
+}
+
+void check(int i, long long result, long long expect)
+{
+    if (result != expect) {
+        printf("ERROR(%d): 0x%04llx != 0x%04llx\n", i, result, expect);
+        err++;
+    }
+}
+
+#define TEST_BREV_LOAD(SZ, TYPE, BUF, SHIFT, EXP) \
+    do { \
+        p = BUF; \
+        for (i = 0; i < SIZE; i++) { \
+            TYPE result; \
+            BREV_LOAD_##SZ(result, p, 1 << (SHIFT - NBITS)); \
+            check(i, result, EXP); \
+        } \
+    } while (0)
+
+#define TEST_BREV_STORE(SZ, TYPE, BUF, VAL, SHIFT) \
+    do { \
+        p = BUF; \
+        memset(BUF, 0xff, sizeof(BUF)); \
+        for (i = 0; i < SIZE; i++) { \
+            BREV_STORE_##SZ(p, (TYPE)(VAL), 1 << (SHIFT - NBITS)); \
+        } \
+        for (i = 0; i < SIZE; i++) { \
+            check(i, BUF[i], bitreverse(i)); \
+        } \
+    } while (0)
+
+#define TEST_BREV_STORE_NEW(SZ, BUF, SHIFT) \
+    do { \
+        p = BUF; \
+        memset(BUF, 0xff, sizeof(BUF)); \
+        for (i = 0; i < SIZE; i++) { \
+            BREV_STORE_##SZ(p, i, 1 << (SHIFT - NBITS)); \
+        } \
+        for (i = 0; i < SIZE; i++) { \
+            check(i, BUF[i], bitreverse(i)); \
+        } \
+    } while (0)
+
+/*
+ * We'll set high_half[i] = i << 16 for use in the .H form of store
+ * which stores from the high half of the word.
+ */
+int high_half[SIZE];
+
+int main()
+{
+    void *p;
+    int i;
+
+    for (i = 0; i < SIZE; i++) {
+        bbuf[i] = bitreverse(i);
+        hbuf[i] = bitreverse(i);
+        wbuf[i] = bitreverse(i);
+        dbuf[i] = bitreverse(i);
+        high_half[i] = i << 16;
+    }
+
+    TEST_BREV_LOAD(b,  int,       bbuf, 16, sext8(i));
+    TEST_BREV_LOAD(ub, int,       bbuf, 16, i);
+    TEST_BREV_LOAD(h,  int,       hbuf, 15, i);
+    TEST_BREV_LOAD(uh, int,       hbuf, 15, i);
+    TEST_BREV_LOAD(w,  int,       wbuf, 14, i);
+    TEST_BREV_LOAD(d,  long long, dbuf, 13, i);
+
+    TEST_BREV_STORE(b, int,       bbuf, i,            16);
+    TEST_BREV_STORE(h, int,       hbuf, i,            15);
+    TEST_BREV_STORE(f, int,       hbuf, high_half[i], 15);
+    TEST_BREV_STORE(w, int,       wbuf, i,            14);
+    TEST_BREV_STORE(d, long long, dbuf, i,            13);
+
+    TEST_BREV_STORE_NEW(bnew, bbuf, 16);
+    TEST_BREV_STORE_NEW(hnew, hbuf, 15);
+    TEST_BREV_STORE_NEW(wnew, wbuf, 14);
+
+    puts(err ? "FAIL" : "PASS");
+    return err ? 1 : 0;
+}
diff --git a/tests/tcg/hexagon/circ.c b/tests/tcg/hexagon/circ.c
new file mode 100644
index 0000000..67a1aa3
--- /dev/null
+++ b/tests/tcg/hexagon/circ.c
@@ -0,0 +1,486 @@
+/*
+ *  Copyright(c) 2019-2021 Qualcomm Innovation Center, Inc. All Rights Reserved.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <stdio.h>
+
+#define DEBUG          0
+#define DEBUG_PRINTF(...) \
+    do { \
+        if (DEBUG) { \
+            printf(__VA_ARGS__); \
+        } \
+    } while (0)
+
+
+#define NBYTES         (1 << 8)
+#define NHALFS         (NBYTES / sizeof(short))
+#define NWORDS         (NBYTES / sizeof(int))
+#define NDOBLS         (NBYTES / sizeof(long long))
+
+long long     dbuf[NDOBLS] __attribute__((aligned(1 << 12))) = {0};
+int           wbuf[NWORDS] __attribute__((aligned(1 << 12))) = {0};
+short         hbuf[NHALFS] __attribute__((aligned(1 << 12))) = {0};
+unsigned char bbuf[NBYTES] __attribute__((aligned(1 << 12))) = {0};
+
+/*
+ * We use the C preporcessor to deal with the combinations of types
+ */
+
+#define INIT(BUF, N) \
+    void init_##BUF(void) \
+    { \
+        int i; \
+        for (i = 0; i < N; i++) { \
+            BUF[i] = i; \
+        } \
+    } \
+
+INIT(bbuf, NBYTES)
+INIT(hbuf, NHALFS)
+INIT(wbuf, NWORDS)
+INIT(dbuf, NDOBLS)
+
+/*
+ * Macros for performing circular load
+ *     RES         result
+ *     ADDR        address
+ *     START       start address of buffer
+ *     LEN         length of buffer (in bytes)
+ *     INC         address increment (in bytes for IMM, elements for REG)
+ */
+#define CIRC_LOAD_IMM(SIZE, RES, ADDR, START, LEN, INC) \
+    __asm__( \
+        "r4 = %3\n\t" \
+        "m0 = r4\n\t" \
+        "cs0 = %2\n\t" \
+        "%0 = mem" #SIZE "(%1++#" #INC ":circ(M0))\n\t" \
+        : "=r"(RES), "+r"(ADDR) \
+        : "r"(START), "r"(LEN) \
+        : "r4", "m0", "cs0")
+#define CIRC_LOAD_IMM_b(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_IMM(b, RES, ADDR, START, LEN, INC)
+#define CIRC_LOAD_IMM_ub(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_IMM(ub, RES, ADDR, START, LEN, INC)
+#define CIRC_LOAD_IMM_h(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_IMM(h, RES, ADDR, START, LEN, INC)
+#define CIRC_LOAD_IMM_uh(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_IMM(uh, RES, ADDR, START, LEN, INC)
+#define CIRC_LOAD_IMM_w(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_IMM(w, RES, ADDR, START, LEN, INC)
+#define CIRC_LOAD_IMM_d(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_IMM(d, RES, ADDR, START, LEN, INC)
+
+/*
+ * The mreg has the following pieces
+ *     mreg[31:28]              increment[10:7]
+ *     mreg[27:24]              K value (used Hexagon v3 and earlier)
+ *     mreg[23:17]              increment[6:0]
+ *     mreg[16:0]               circular buffer length
+ */
+static int build_mreg(int inc, int K, int len)
+{
+    return ((inc & 0x780) << 21) |
+           ((K & 0xf) << 24) |
+           ((inc & 0x7f) << 17) |
+           (len & 0x1ffff);
+}
+
+#define CIRC_LOAD_REG(SIZE, RES, ADDR, START, LEN, INC) \
+    __asm__( \
+        "r4 = %2\n\t" \
+        "m1 = r4\n\t" \
+        "cs1 = %3\n\t" \
+        "%0 = mem" #SIZE "(%1++I:circ(M1))\n\t" \
+        : "=r"(RES), "+r"(ADDR) \
+        : "r"(build_mreg((INC), 0, (LEN))), \
+          "r"(START) \
+        : "r4", "m1", "cs1")
+#define CIRC_LOAD_REG_b(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_REG(b, RES, ADDR, START, LEN, INC)
+#define CIRC_LOAD_REG_ub(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_REG(ub, RES, ADDR, START, LEN, INC)
+#define CIRC_LOAD_REG_h(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_REG(h, RES, ADDR, START, LEN, INC)
+#define CIRC_LOAD_REG_uh(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_REG(uh, RES, ADDR, START, LEN, INC)
+#define CIRC_LOAD_REG_w(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_REG(w, RES, ADDR, START, LEN, INC)
+#define CIRC_LOAD_REG_d(RES, ADDR, START, LEN, INC) \
+    CIRC_LOAD_REG(d, RES, ADDR, START, LEN, INC)
+
+/*
+ * Macros for performing circular store
+ *     VAL         value to store
+ *     ADDR        address
+ *     START       start address of buffer
+ *     LEN         length of buffer (in bytes)
+ *     INC         address increment (in bytes for IMM, elements for REG)
+ */
+#define CIRC_STORE_IMM(SIZE, PART, VAL, ADDR, START, LEN, INC) \
+    __asm__( \
+        "r4 = %3\n\t" \
+        "m0 = r4\n\t" \
+        "cs0 = %1\n\t" \
+        "mem" #SIZE "(%0++#" #INC ":circ(M0)) = %2" PART "\n\t" \
+        : "+r"(ADDR) \
+        : "r"(START), "r"(VAL), "r"(LEN) \
+        : "r4", "m0", "cs0", "memory")
+#define CIRC_STORE_IMM_b(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_IMM(b, "", VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_IMM_h(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_IMM(h, "", VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_IMM_f(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_IMM(h, ".H", VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_IMM_w(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_IMM(w, "", VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_IMM_d(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_IMM(d, "", VAL, ADDR, START, LEN, INC)
+
+#define CIRC_STORE_NEW_IMM(SIZE, VAL, ADDR, START, LEN, INC) \
+    __asm__( \
+        "r4 = %3\n\t" \
+        "m0 = r4\n\t" \
+        "cs0 = %1\n\t" \
+        "{\n\t" \
+        "    r5 = %2\n\t" \
+        "    mem" #SIZE "(%0++#" #INC ":circ(M0)) = r5.new\n\t" \
+        "}\n\t" \
+        : "+r"(ADDR) \
+        : "r"(START), "r"(VAL), "r"(LEN) \
+        : "r4", "r5", "m0", "cs0", "memory")
+#define CIRC_STORE_IMM_bnew(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_NEW_IMM(b, VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_IMM_hnew(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_NEW_IMM(h, VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_IMM_wnew(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_NEW_IMM(w, VAL, ADDR, START, LEN, INC)
+
+#define CIRC_STORE_REG(SIZE, PART, VAL, ADDR, START, LEN, INC) \
+    __asm__( \
+        "r4 = %1\n\t" \
+        "m1 = r4\n\t" \
+        "cs1 = %2\n\t" \
+        "mem" #SIZE "(%0++I:circ(M1)) = %3" PART "\n\t" \
+        : "+r"(ADDR) \
+        : "r"(build_mreg((INC), 0, (LEN))), \
+          "r"(START), \
+          "r"(VAL) \
+        : "r4", "m1", "cs1", "memory")
+#define CIRC_STORE_REG_b(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_REG(b, "", VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_REG_h(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_REG(h, "", VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_REG_f(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_REG(h, ".H", VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_REG_w(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_REG(w, "", VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_REG_d(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_REG(d, "", VAL, ADDR, START, LEN, INC)
+
+#define CIRC_STORE_NEW_REG(SIZE, VAL, ADDR, START, LEN, INC) \
+    __asm__( \
+        "r4 = %1\n\t" \
+        "m1 = r4\n\t" \
+        "cs1 = %2\n\t" \
+        "{\n\t" \
+        "    r5 = %3\n\t" \
+        "    mem" #SIZE "(%0++I:circ(M1)) = r5.new\n\t" \
+        "}\n\t" \
+        : "+r"(ADDR) \
+        : "r"(build_mreg((INC), 0, (LEN))), \
+          "r"(START), \
+          "r"(VAL) \
+        : "r4", "r5", "m1", "cs1", "memory")
+#define CIRC_STORE_REG_bnew(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_NEW_REG(b, VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_REG_hnew(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_NEW_REG(h, VAL, ADDR, START, LEN, INC)
+#define CIRC_STORE_REG_wnew(VAL, ADDR, START, LEN, INC) \
+    CIRC_STORE_NEW_REG(w, VAL, ADDR, START, LEN, INC)
+
+
+int err;
+
+/* We'll test increments +1 and -1 */
+void check_load(int i, long long result, int inc, int size)
+{
+    int expect = (i * inc);
+    while (expect >= size) {
+        expect -= size;
+    }
+    while (expect < 0) {
+        expect += size;
+    }
+    if (result != expect) {
+        printf("ERROR(%d): %lld != %d\n", i, result, expect);
+        err++;
+    }
+}
+
+#define TEST_LOAD_IMM(SZ, TYPE, BUF, BUFSIZE, INC, FMT) \
+void circ_test_load_imm_##SZ(void) \
+{ \
+    TYPE *p = (TYPE *)BUF; \
+    int size = 10; \
+    int i; \
+    for (i = 0; i < BUFSIZE; i++) { \
+        TYPE element; \
+        CIRC_LOAD_IMM_##SZ(element, p, BUF, size * sizeof(TYPE), (INC)); \
+        DEBUG_PRINTF("i = %2d, p = 0x%p, element = %2" #FMT "\n", \
+                     i, p, element); \
+        check_load(i, element, ((INC) / (int)sizeof(TYPE)), size); \
+    } \
+    p = (TYPE *)BUF; \
+    for (i = 0; i < BUFSIZE; i++) { \
+        TYPE element; \
+        CIRC_LOAD_IMM_##SZ(element, p, BUF, size * sizeof(TYPE), -(INC)); \
+        DEBUG_PRINTF("i = %2d, p = 0x%p, element = %2" #FMT "\n", \
+                     i, p, element); \
+        check_load(i, element, (-(INC) / (int)sizeof(TYPE)), size); \
+    } \
+}
+
+TEST_LOAD_IMM(b,  char,           bbuf, NBYTES, 1, d)
+TEST_LOAD_IMM(ub, unsigned char,  bbuf, NBYTES, 1, d)
+TEST_LOAD_IMM(h,  short,          hbuf, NHALFS, 2, d)
+TEST_LOAD_IMM(uh, unsigned short, hbuf, NHALFS, 2, d)
+TEST_LOAD_IMM(w,  int,            wbuf, NWORDS, 4, d)
+TEST_LOAD_IMM(d,  long long,      dbuf, NDOBLS, 8, lld)
+
+#define TEST_LOAD_REG(SZ, TYPE, BUF, BUFSIZE, FMT) \
+void circ_test_load_reg_##SZ(void) \
+{ \
+    TYPE *p = (TYPE *)BUF; \
+    int size = 13; \
+    int i; \
+    for (i = 0; i < BUFSIZE; i++) { \
+        TYPE element; \
+        CIRC_LOAD_REG_##SZ(element, p, BUF, size * sizeof(TYPE), 1); \
+        DEBUG_PRINTF("i = %2d, p = 0x%p, element = %2" #FMT "\n", \
+                     i, p, element); \
+        check_load(i, element, 1, size); \
+    } \
+    p = (TYPE *)BUF; \
+    for (i = 0; i < BUFSIZE; i++) { \
+        TYPE element; \
+        CIRC_LOAD_REG_##SZ(element, p, BUF, size * sizeof(TYPE), -1); \
+        DEBUG_PRINTF("i = %2d, p = 0x%p, element = %2" #FMT "\n", \
+                     i, p, element); \
+        check_load(i, element, -1, size); \
+    } \
+}
+
+TEST_LOAD_REG(b,  char,           bbuf, NBYTES, d)
+TEST_LOAD_REG(ub, unsigned char,  bbuf, NBYTES, d)
+TEST_LOAD_REG(h,  short,          hbuf, NHALFS, d)
+TEST_LOAD_REG(uh, unsigned short, hbuf, NHALFS, d)
+TEST_LOAD_REG(w,  int,            wbuf, NWORDS, d)
+TEST_LOAD_REG(d,  long long,      dbuf, NDOBLS, lld)
+
+/* The circular stores will wrap around somewhere inside the buffer */
+#define CIRC_VAL(SZ, TYPE, BUFSIZE) \
+TYPE circ_val_##SZ(int i, int inc, int size) \
+{ \
+    int mod = BUFSIZE % size; \
+    int elem = i * inc; \
+    if (elem < 0) { \
+        if (-elem <= size - mod) { \
+            return (elem + BUFSIZE - mod); \
+        } else { \
+            return (elem + BUFSIZE + size - mod); \
+        } \
+    } else if (elem < mod) {\
+        return (elem + BUFSIZE - mod); \
+    } else { \
+        return (elem + BUFSIZE - size - mod); \
+    } \
+}
+
+CIRC_VAL(b, unsigned char, NBYTES)
+CIRC_VAL(h, short,         NHALFS)
+CIRC_VAL(w, int,           NWORDS)
+CIRC_VAL(d, long long,     NDOBLS)
+
+/*
+ * Circular stores should only write to the first "size" elements of the buffer
+ * the remainder of the elements should have BUF[i] == i
+ */
+#define CHECK_STORE(SZ, BUF, BUFSIZE, FMT) \
+void check_store_##SZ(int inc, int size) \
+{ \
+    int i; \
+    for (i = 0; i < size; i++) { \
+        DEBUG_PRINTF(#BUF "[%3d] = 0x%02" #FMT ", guess = 0x%02" #FMT "\n", \
+                     i, BUF[i], circ_val_##SZ(i, inc, size)); \
+        if (BUF[i] != circ_val_##SZ(i, inc, size)) { \
+            printf("ERROR(%3d): 0x%02" #FMT " != 0x%02" #FMT "\n", \
+                   i, BUF[i], circ_val_##SZ(i, inc, size)); \
+            err++; \
+        } \
+    } \
+    for (i = size; i < BUFSIZE; i++) { \
+        if (BUF[i] != i) { \
+            printf("ERROR(%3d): 0x%02" #FMT " != 0x%02x\n", i, BUF[i], i); \
+            err++; \
+        } \
+    } \
+}
+
+CHECK_STORE(b, bbuf, NBYTES, x)
+CHECK_STORE(h, hbuf, NHALFS, x)
+CHECK_STORE(w, wbuf, NWORDS, x)
+CHECK_STORE(d, dbuf, NDOBLS, llx)
+
+#define CIRC_TEST_STORE_IMM(SZ, CHK, TYPE, BUF, BUFSIZE, SHIFT, INC) \
+void circ_test_store_imm_##SZ(void) \
+{ \
+    unsigned int size = 27; \
+    TYPE *p = BUF; \
+    TYPE val = 0; \
+    int i; \
+    init_##BUF(); \
+    for (i = 0; i < BUFSIZE; i++) { \
+        CIRC_STORE_IMM_##SZ(val << SHIFT, p, BUF, size * sizeof(TYPE), INC); \
+        val++; \
+    } \
+    check_store_##CHK(((INC) / (int)sizeof(TYPE)), size); \
+    p = BUF; \
+    val = 0; \
+    init_##BUF(); \
+    for (i = 0; i < BUFSIZE; i++) { \
+        CIRC_STORE_IMM_##SZ(val << SHIFT, p, BUF, size * sizeof(TYPE), \
+                            -(INC)); \
+        val++; \
+    } \
+    check_store_##CHK((-(INC) / (int)sizeof(TYPE)), size); \
+}
+
+CIRC_TEST_STORE_IMM(b,    b, unsigned char, bbuf, NBYTES, 0,  1)
+CIRC_TEST_STORE_IMM(h,    h, short,         hbuf, NHALFS, 0,  2)
+CIRC_TEST_STORE_IMM(f,    h, short,         hbuf, NHALFS, 16, 2)
+CIRC_TEST_STORE_IMM(w,    w, int,           wbuf, NWORDS, 0,  4)
+CIRC_TEST_STORE_IMM(d,    d, long long,     dbuf, NDOBLS, 0,  8)
+CIRC_TEST_STORE_IMM(bnew, b, unsigned char, bbuf, NBYTES, 0,  1)
+CIRC_TEST_STORE_IMM(hnew, h, short,         hbuf, NHALFS, 0,  2)
+CIRC_TEST_STORE_IMM(wnew, w, int,           wbuf, NWORDS, 0,  4)
+
+#define CIRC_TEST_STORE_REG(SZ, CHK, TYPE, BUF, BUFSIZE, SHIFT) \
+void circ_test_store_reg_##SZ(void) \
+{ \
+    TYPE *p = BUF; \
+    unsigned int size = 19; \
+    TYPE val = 0; \
+    int i; \
+    init_##BUF(); \
+    for (i = 0; i < BUFSIZE; i++) { \
+        CIRC_STORE_REG_##SZ(val << SHIFT, p, BUF, size * sizeof(TYPE), 1); \
+        val++; \
+    } \
+    check_store_##CHK(1, size); \
+    p = BUF; \
+    val = 0; \
+    init_##BUF(); \
+    for (i = 0; i < BUFSIZE; i++) { \
+        CIRC_STORE_REG_##SZ(val << SHIFT, p, BUF, size * sizeof(TYPE), -1); \
+        val++; \
+    } \
+    check_store_##CHK(-1, size); \
+}
+
+CIRC_TEST_STORE_REG(b,    b, unsigned char, bbuf, NBYTES, 0)
+CIRC_TEST_STORE_REG(h,    h, short,         hbuf, NHALFS, 0)
+CIRC_TEST_STORE_REG(f,    h, short,         hbuf, NHALFS, 16)
+CIRC_TEST_STORE_REG(w,    w, int,           wbuf, NWORDS, 0)
+CIRC_TEST_STORE_REG(d,    d, long long,     dbuf, NDOBLS, 0)
+CIRC_TEST_STORE_REG(bnew, b, unsigned char, bbuf, NBYTES, 0)
+CIRC_TEST_STORE_REG(hnew, h, short,         hbuf, NHALFS, 0)
+CIRC_TEST_STORE_REG(wnew, w, int,           wbuf, NWORDS, 0)
+
+/* Test the old scheme used in Hexagon V3 */
+static void circ_test_v3(void)
+{
+    int *p = wbuf;
+    int size = 15;
+    int K = 4;      /* 64 bytes */
+    int element;
+    int i;
+
+    init_wbuf();
+
+    for (i = 0; i < NWORDS; i++) {
+        __asm__(
+            "r4 = %2\n\t"
+            "m1 = r4\n\t"
+            "%0 = memw(%1++I:circ(M1))\n\t"
+            : "=r"(element), "+r"(p)
+            : "r"(build_mreg(1, K, size * sizeof(int)))
+            : "r4", "m1");
+        DEBUG_PRINTF("i = %2d, p = 0x%p, element = %2d\n", i, p, element);
+        check_load(i, element, 1, size);
+    }
+}
+
+int main()
+{
+    init_bbuf();
+    init_hbuf();
+    init_wbuf();
+    init_dbuf();
+
+    DEBUG_PRINTF("NBYTES = %d\n", NBYTES);
+    DEBUG_PRINTF("Address of dbuf = 0x%p\n", dbuf);
+    DEBUG_PRINTF("Address of wbuf = 0x%p\n", wbuf);
+    DEBUG_PRINTF("Address of hbuf = 0x%p\n", hbuf);
+    DEBUG_PRINTF("Address of bbuf = 0x%p\n", bbuf);
+
+    circ_test_load_imm_b();
+    circ_test_load_imm_ub();
+    circ_test_load_imm_h();
+    circ_test_load_imm_uh();
+    circ_test_load_imm_w();
+    circ_test_load_imm_d();
+
+    circ_test_load_reg_b();
+    circ_test_load_reg_ub();
+    circ_test_load_reg_h();
+    circ_test_load_reg_uh();
+    circ_test_load_reg_w();
+    circ_test_load_reg_d();
+
+    circ_test_store_imm_b();
+    circ_test_store_imm_h();
+    circ_test_store_imm_f();
+    circ_test_store_imm_w();
+    circ_test_store_imm_d();
+    circ_test_store_imm_bnew();
+    circ_test_store_imm_hnew();
+    circ_test_store_imm_wnew();
+
+    circ_test_store_reg_b();
+    circ_test_store_reg_h();
+    circ_test_store_reg_f();
+    circ_test_store_reg_w();
+    circ_test_store_reg_d();
+    circ_test_store_reg_bnew();
+    circ_test_store_reg_hnew();
+    circ_test_store_reg_wnew();
+
+    circ_test_v3();
+
+    puts(err ? "FAIL" : "PASS");
+    return err ? 1 : 0;
+}
diff --git a/tests/tcg/hexagon/fpstuff.c b/tests/tcg/hexagon/fpstuff.c
index e4f1a0e..0dff429 100644
--- a/tests/tcg/hexagon/fpstuff.c
+++ b/tests/tcg/hexagon/fpstuff.c
@@ -37,10 +37,12 @@ const int SF_NaN =                        0x7fc00000;
 const int SF_NaN_special =                0x7f800001;
 const int SF_ANY =                        0x3f800000;
 const int SF_HEX_NAN =                    0xffffffff;
+const int SF_small_neg =                  0xab98fba8;
 
 const long long DF_NaN =                  0x7ff8000000000000ULL;
 const long long DF_ANY =                  0x3f80000000000000ULL;
 const long long DF_HEX_NAN =              0xffffffffffffffffULL;
+const long long DF_small_neg =            0xbd731f7500000000ULL;
 
 int err;
 
@@ -248,6 +250,87 @@ static void check_dfminmax(void)
     check_fpstatus(usr, FPINVF);
 }
 
+static void check_recip_exception(void)
+{
+    int result;
+    int usr;
+
+    /*
+     * Check that sfrecipa doesn't set status bits when
+     * a NaN with bit 22 non-zero is passed
+     */
+    asm (CLEAR_FPSTATUS
+         "%0,p0 = sfrecipa(%2, %3)\n\t"
+         "%1 = usr\n\t"
+         : "=r"(result), "=r"(usr) : "r"(SF_NaN), "r"(SF_ANY)
+         : "r2", "p0", "usr");
+    check32(result, SF_HEX_NAN);
+    check_fpstatus(usr, 0);
+
+    asm (CLEAR_FPSTATUS
+         "%0,p0 = sfrecipa(%2, %3)\n\t"
+         "%1 = usr\n\t"
+         : "=r"(result), "=r"(usr) : "r"(SF_ANY), "r"(SF_NaN)
+         : "r2", "p0", "usr");
+    check32(result, SF_HEX_NAN);
+    check_fpstatus(usr, 0);
+
+    asm (CLEAR_FPSTATUS
+         "%0,p0 = sfrecipa(%2, %2)\n\t"
+         "%1 = usr\n\t"
+         : "=r"(result), "=r"(usr) : "r"(SF_NaN)
+         : "r2", "p0", "usr");
+    check32(result, SF_HEX_NAN);
+    check_fpstatus(usr, 0);
+
+    /*
+     * Check that sfrecipa doesn't set status bits when
+     * a NaN with bit 22 zero is passed
+     */
+    asm (CLEAR_FPSTATUS
+         "%0,p0 = sfrecipa(%2, %3)\n\t"
+         "%1 = usr\n\t"
+         : "=r"(result), "=r"(usr) : "r"(SF_NaN_special), "r"(SF_ANY)
+         : "r2", "p0", "usr");
+    check32(result, SF_HEX_NAN);
+    check_fpstatus(usr, FPINVF);
+
+    asm (CLEAR_FPSTATUS
+         "%0,p0 = sfrecipa(%2, %3)\n\t"
+         "%1 = usr\n\t"
+         : "=r"(result), "=r"(usr) : "r"(SF_ANY), "r"(SF_NaN_special)
+         : "r2", "p0", "usr");
+    check32(result, SF_HEX_NAN);
+    check_fpstatus(usr, FPINVF);
+
+    asm (CLEAR_FPSTATUS
+         "%0,p0 = sfrecipa(%2, %2)\n\t"
+         "%1 = usr\n\t"
+         : "=r"(result), "=r"(usr) : "r"(SF_NaN_special)
+         : "r2", "p0", "usr");
+    check32(result, SF_HEX_NAN);
+    check_fpstatus(usr, FPINVF);
+
+    /*
+     * Check that sfrecipa properly sets divid-by-zero
+     */
+        asm (CLEAR_FPSTATUS
+         "%0,p0 = sfrecipa(%2, %3)\n\t"
+         "%1 = usr\n\t"
+         : "=r"(result), "=r"(usr) : "r"(0x885dc960), "r"(0x80000000)
+         : "r2", "p0", "usr");
+    check32(result, 0x3f800000);
+    check_fpstatus(usr, FPDBZF);
+
+    asm (CLEAR_FPSTATUS
+         "%0,p0 = sfrecipa(%2, %3)\n\t"
+         "%1 = usr\n\t"
+         : "=r"(result), "=r"(usr) : "r"(0x7f800000), "r"(SF_ZERO)
+         : "r2", "p0", "usr");
+    check32(result, 0x3f800000);
+    check_fpstatus(usr, 0);
+}
+
 static void check_canonical_NaN(void)
 {
     int sf_result;
@@ -358,12 +441,171 @@ static void check_canonical_NaN(void)
     check_fpstatus(usr, 0);
 }
 
+static void check_invsqrta(void)
+{
+    int result;
+    int predval;
+
+    asm volatile("%0,p0 = sfinvsqrta(%2)\n\t"
+                 "%1 = p0\n\t"
+                 : "+r"(result), "=r"(predval)
+                 : "r"(0x7f800000)
+                 : "p0");
+    check32(result, 0xff800000);
+    check32(predval, 0x0);
+}
+
+static void check_float2int_convs()
+{
+    int res32;
+    long long res64;
+    int usr;
+
+    /*
+     * Check that the various forms of float-to-unsigned
+     *  check sign before rounding
+     */
+        asm(CLEAR_FPSTATUS
+        "%0 = convert_sf2uw(%2)\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res32), "=r"(usr) : "r"(SF_small_neg)
+        : "r2", "usr");
+    check32(res32, 0);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_sf2uw(%2):chop\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res32), "=r"(usr) : "r"(SF_small_neg)
+        : "r2", "usr");
+    check32(res32, 0);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_sf2ud(%2)\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res64), "=r"(usr) : "r"(SF_small_neg)
+        : "r2", "usr");
+    check64(res64, 0);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_sf2ud(%2):chop\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res64), "=r"(usr) : "r"(SF_small_neg)
+        : "r2", "usr");
+    check64(res64, 0);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_df2uw(%2)\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res32), "=r"(usr) : "r"(DF_small_neg)
+        : "r2", "usr");
+    check32(res32, 0);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_df2uw(%2):chop\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res32), "=r"(usr) : "r"(DF_small_neg)
+        : "r2", "usr");
+    check32(res32, 0);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_df2ud(%2)\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res64), "=r"(usr) : "r"(DF_small_neg)
+        : "r2", "usr");
+    check64(res64, 0);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_df2ud(%2):chop\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res64), "=r"(usr) : "r"(DF_small_neg)
+        : "r2", "usr");
+    check64(res64, 0);
+    check_fpstatus(usr, FPINVF);
+
+    /*
+     * Check that the various forms of float-to-signed return -1 for NaN
+     */
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_sf2w(%2)\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res32), "=r"(usr) : "r"(SF_NaN)
+        : "r2", "usr");
+    check32(res32, -1);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_sf2w(%2):chop\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res32), "=r"(usr) : "r"(SF_NaN)
+        : "r2", "usr");
+    check32(res32, -1);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_sf2d(%2)\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res64), "=r"(usr) : "r"(SF_NaN)
+        : "r2", "usr");
+    check64(res64, -1);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_sf2d(%2):chop\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res64), "=r"(usr) : "r"(SF_NaN)
+        : "r2", "usr");
+    check64(res64, -1);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_df2w(%2)\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res32), "=r"(usr) : "r"(DF_NaN)
+        : "r2", "usr");
+    check32(res32, -1);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_df2w(%2):chop\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res32), "=r"(usr) : "r"(DF_NaN)
+        : "r2", "usr");
+    check32(res32, -1);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_df2d(%2)\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res64), "=r"(usr) : "r"(DF_NaN)
+        : "r2", "usr");
+    check64(res64, -1);
+    check_fpstatus(usr, FPINVF);
+
+    asm(CLEAR_FPSTATUS
+        "%0 = convert_df2d(%2):chop\n\t"
+        "%1 = usr\n\t"
+        : "=r"(res64), "=r"(usr) : "r"(DF_NaN)
+        : "r2", "usr");
+    check64(res64, -1);
+    check_fpstatus(usr, FPINVF);
+}
+
 int main()
 {
     check_compare_exception();
     check_sfminmax();
     check_dfminmax();
+    check_recip_exception();
     check_canonical_NaN();
+    check_invsqrta();
+    check_float2int_convs();
 
     puts(err ? "FAIL" : "PASS");
     return err ? 1 : 0;
diff --git a/tests/tcg/hexagon/load_align.c b/tests/tcg/hexagon/load_align.c
new file mode 100644
index 0000000..12fc9cb
--- /dev/null
+++ b/tests/tcg/hexagon/load_align.c
@@ -0,0 +1,415 @@
+/*
+ *  Copyright(c) 2019-2021 Qualcomm Innovation Center, Inc. All Rights Reserved.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+/*
+ * Test load align instructions
+ *
+ * Example
+ *     r1:0 = memh_fifo(r1+#0)
+ * loads a half word from memory, shifts the destination register
+ * right by one half word and inserts the loaded value into the high
+ * half word of the destination.
+ *
+ * There are 8 addressing modes and byte and half word variants, for a
+ * total of 16 instructions to test
+ */
+
+#include <stdio.h>
+#include <string.h>
+
+int err;
+
+char buf[16] __attribute__((aligned(1 << 16)));
+
+void init_buf(void)
+{
+    int i;
+    for (i = 0; i < 16; i++) {
+        buf[i] = i + 1;
+    }
+}
+
+void __check(int line, long long result, long long expect)
+{
+    if (result != expect) {
+        printf("ERROR at line %d: 0x%016llx != 0x%016llx\n",
+               line, result, expect);
+        err++;
+    }
+}
+
+#define check(RES, EXP) __check(__LINE__, RES, EXP)
+
+void __checkp(int line, void *p, void *expect)
+{
+    if (p != expect) {
+        printf("ERROR at line %d: 0x%p != 0x%p\n", line, p, expect);
+        err++;
+    }
+}
+
+#define checkp(RES, EXP) __checkp(__LINE__, RES, EXP)
+
+/*
+ ****************************************************************************
+ * _io addressing mode (addr + offset)
+ */
+#define LOAD_io(SZ, RES, ADDR, OFF) \
+    __asm__( \
+        "%0 = mem" #SZ "_fifo(%1+#" #OFF ")\n\t" \
+        : "+r"(RES) \
+        : "r"(ADDR))
+#define LOAD_io_b(RES, ADDR, OFF) \
+    LOAD_io(b, RES, ADDR, OFF)
+#define LOAD_io_h(RES, ADDR, OFF) \
+    LOAD_io(h, RES, ADDR, OFF)
+
+#define TEST_io(NAME, SZ, SIZE, EXP1, EXP2, EXP3, EXP4) \
+void test_##NAME(void) \
+{ \
+    long long result = ~0LL; \
+    LOAD_io_##SZ(result, buf, 0 * (SIZE)); \
+    check(result, (EXP1)); \
+    LOAD_io_##SZ(result, buf, 1 * (SIZE)); \
+    check(result, (EXP2)); \
+    LOAD_io_##SZ(result, buf, 2 * (SIZE)); \
+    check(result, (EXP3)); \
+    LOAD_io_##SZ(result, buf, 3 * (SIZE)); \
+    check(result, (EXP4)); \
+}
+
+TEST_io(loadalignb_io, b, 1,
+        0x01ffffffffffffffLL, 0x0201ffffffffffffLL,
+        0x030201ffffffffffLL, 0x04030201ffffffffLL)
+TEST_io(loadalignh_io, h, 2,
+        0x0201ffffffffffffLL, 0x04030201ffffffffLL,
+        0x060504030201ffffLL, 0x0807060504030201LL)
+
+/*
+ ****************************************************************************
+ * _ur addressing mode (index << offset + base)
+ */
+#define LOAD_ur(SZ, RES, SHIFT, IDX) \
+    __asm__( \
+        "%0 = mem" #SZ "_fifo(%1<<#" #SHIFT " + ##buf)\n\t" \
+        : "+r"(RES) \
+        : "r"(IDX))
+#define LOAD_ur_b(RES, SHIFT, IDX) \
+    LOAD_ur(b, RES, SHIFT, IDX)
+#define LOAD_ur_h(RES, SHIFT, IDX) \
+    LOAD_ur(h, RES, SHIFT, IDX)
+
+#define TEST_ur(NAME, SZ, SHIFT, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    long long result = ~0LL; \
+    LOAD_ur_##SZ(result, (SHIFT), 0); \
+    check(result, (RES1)); \
+    LOAD_ur_##SZ(result, (SHIFT), 1); \
+    check(result, (RES2)); \
+    LOAD_ur_##SZ(result, (SHIFT), 2); \
+    check(result, (RES3)); \
+    LOAD_ur_##SZ(result, (SHIFT), 3); \
+    check(result, (RES4)); \
+}
+
+TEST_ur(loadalignb_ur, b, 1,
+        0x01ffffffffffffffLL, 0x0301ffffffffffffLL,
+        0x050301ffffffffffLL, 0x07050301ffffffffLL)
+TEST_ur(loadalignh_ur, h, 1,
+        0x0201ffffffffffffLL, 0x04030201ffffffffLL,
+        0x060504030201ffffLL, 0x0807060504030201LL)
+
+/*
+ ****************************************************************************
+ * _ap addressing mode (addr = base)
+ */
+#define LOAD_ap(SZ, RES, PTR, ADDR) \
+    __asm__(  \
+        "%0 = mem" #SZ "_fifo(%1 = ##" #ADDR ")\n\t" \
+        : "+r"(RES), "=r"(PTR))
+#define LOAD_ap_b(RES, PTR, ADDR) \
+    LOAD_ap(b, RES, PTR, ADDR)
+#define LOAD_ap_h(RES, PTR, ADDR) \
+    LOAD_ap(h, RES, PTR, ADDR)
+
+#define TEST_ap(NAME, SZ, SIZE, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    long long result = ~0LL; \
+    void *ptr; \
+    LOAD_ap_##SZ(result, ptr, (buf + 0 * (SIZE))); \
+    check(result, (RES1)); \
+    checkp(ptr, &buf[0 * (SIZE)]); \
+    LOAD_ap_##SZ(result, ptr, (buf + 1 * (SIZE))); \
+    check(result, (RES2)); \
+    checkp(ptr, &buf[1 * (SIZE)]); \
+    LOAD_ap_##SZ(result, ptr, (buf + 2 * (SIZE))); \
+    check(result, (RES3)); \
+    checkp(ptr, &buf[2 * (SIZE)]); \
+    LOAD_ap_##SZ(result, ptr, (buf + 3 * (SIZE))); \
+    check(result, (RES4)); \
+    checkp(ptr, &buf[3 * (SIZE)]); \
+}
+
+TEST_ap(loadalignb_ap, b, 1,
+        0x01ffffffffffffffLL, 0x0201ffffffffffffLL,
+        0x030201ffffffffffLL, 0x04030201ffffffffLL)
+TEST_ap(loadalignh_ap, h, 2,
+        0x0201ffffffffffffLL, 0x04030201ffffffffLL,
+        0x060504030201ffffLL, 0x0807060504030201LL)
+
+/*
+ ****************************************************************************
+ * _rp addressing mode (addr ++ modifer-reg)
+ */
+#define LOAD_pr(SZ, RES, PTR, INC) \
+    __asm__( \
+        "m0 = %2\n\t" \
+        "%0 = mem" #SZ "_fifo(%1++m0)\n\t" \
+        : "+r"(RES), "+r"(PTR) \
+        : "r"(INC) \
+        : "m0")
+#define LOAD_pr_b(RES, PTR, INC) \
+    LOAD_pr(b, RES, PTR, INC)
+#define LOAD_pr_h(RES, PTR, INC) \
+    LOAD_pr(h, RES, PTR, INC)
+
+#define TEST_pr(NAME, SZ, SIZE, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    long long result = ~0LL; \
+    void *ptr = buf; \
+    LOAD_pr_##SZ(result, ptr, (SIZE)); \
+    check(result, (RES1)); \
+    checkp(ptr, &buf[1 * (SIZE)]); \
+    LOAD_pr_##SZ(result, ptr, (SIZE)); \
+    check(result, (RES2)); \
+    checkp(ptr, &buf[2 * (SIZE)]); \
+    LOAD_pr_##SZ(result, ptr, (SIZE)); \
+    check(result, (RES3)); \
+    checkp(ptr, &buf[3 * (SIZE)]); \
+    LOAD_pr_##SZ(result, ptr, (SIZE)); \
+    check(result, (RES4)); \
+    checkp(ptr, &buf[4 * (SIZE)]); \
+}
+
+TEST_pr(loadalignb_pr, b, 1,
+        0x01ffffffffffffffLL, 0x0201ffffffffffffLL,
+        0x030201ffffffffffLL, 0x04030201ffffffffLL)
+TEST_pr(loadalignh_pr, h, 2,
+        0x0201ffffffffffffLL, 0x04030201ffffffffLL,
+        0x060504030201ffffLL, 0x0807060504030201LL)
+
+/*
+ ****************************************************************************
+ * _pbr addressing mode (addr ++ modifer-reg:brev)
+ */
+#define LOAD_pbr(SZ, RES, PTR) \
+    __asm__( \
+        "r4 = #(1 << (16 - 3))\n\t" \
+        "m0 = r4\n\t" \
+        "%0 = mem" #SZ "_fifo(%1++m0:brev)\n\t" \
+        : "+r"(RES), "+r"(PTR) \
+        : \
+        : "r4", "m0")
+#define LOAD_pbr_b(RES, PTR) \
+    LOAD_pbr(b, RES, PTR)
+#define LOAD_pbr_h(RES, PTR) \
+    LOAD_pbr(h, RES, PTR)
+
+#define TEST_pbr(NAME, SZ, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    long long result = ~0LL; \
+    void *ptr = buf; \
+    LOAD_pbr_##SZ(result, ptr); \
+    check(result, (RES1)); \
+    LOAD_pbr_##SZ(result, ptr); \
+    check(result, (RES2)); \
+    LOAD_pbr_##SZ(result, ptr); \
+    check(result, (RES3)); \
+    LOAD_pbr_##SZ(result, ptr); \
+    check(result, (RES4)); \
+}
+
+TEST_pbr(loadalignb_pbr, b,
+    0x01ffffffffffffffLL, 0x0501ffffffffffffLL,
+    0x030501ffffffffffLL, 0x07030501ffffffffLL)
+TEST_pbr(loadalignh_pbr, h,
+    0x0201ffffffffffffLL, 0x06050201ffffffffLL,
+    0x040306050201ffffLL, 0x0807040306050201LL)
+
+/*
+ ****************************************************************************
+ * _pi addressing mode (addr ++ inc)
+ */
+#define LOAD_pi(SZ, RES, PTR, INC) \
+    __asm__( \
+        "%0 = mem" #SZ "_fifo(%1++#" #INC ")\n\t" \
+        : "+r"(RES), "+r"(PTR))
+#define LOAD_pi_b(RES, PTR, INC) \
+    LOAD_pi(b, RES, PTR, INC)
+#define LOAD_pi_h(RES, PTR, INC) \
+    LOAD_pi(h, RES, PTR, INC)
+
+#define TEST_pi(NAME, SZ, INC, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    long long result = ~0LL; \
+    void *ptr = buf; \
+    LOAD_pi_##SZ(result, ptr, (INC)); \
+    check(result, (RES1)); \
+    checkp(ptr, &buf[1 * (INC)]); \
+    LOAD_pi_##SZ(result, ptr, (INC)); \
+    check(result, (RES2)); \
+    checkp(ptr, &buf[2 * (INC)]); \
+    LOAD_pi_##SZ(result, ptr, (INC)); \
+    check(result, (RES3)); \
+    checkp(ptr, &buf[3 * (INC)]); \
+    LOAD_pi_##SZ(result, ptr, (INC)); \
+    check(result, (RES4)); \
+    checkp(ptr, &buf[4 * (INC)]); \
+}
+
+TEST_pi(loadalignb_pi, b, 1,
+        0x01ffffffffffffffLL, 0x0201ffffffffffffLL,
+        0x030201ffffffffffLL, 0x04030201ffffffffLL)
+TEST_pi(loadalignh_pi, h, 2,
+        0x0201ffffffffffffLL, 0x04030201ffffffffLL,
+        0x060504030201ffffLL, 0x0807060504030201LL)
+
+/*
+ ****************************************************************************
+ * _pci addressing mode (addr ++ inc:circ)
+ */
+#define LOAD_pci(SZ, RES, PTR, START, LEN, INC) \
+    __asm__( \
+        "r4 = %3\n\t" \
+        "m0 = r4\n\t" \
+        "cs0 = %2\n\t" \
+        "%0 = mem" #SZ "_fifo(%1++#" #INC ":circ(m0))\n\t" \
+        : "+r"(RES), "+r"(PTR) \
+        : "r"(START), "r"(LEN) \
+        : "r4", "m0", "cs0")
+#define LOAD_pci_b(RES, PTR, START, LEN, INC) \
+    LOAD_pci(b, RES, PTR, START, LEN, INC)
+#define LOAD_pci_h(RES, PTR, START, LEN, INC) \
+    LOAD_pci(h, RES, PTR, START, LEN, INC)
+
+#define TEST_pci(NAME, SZ, LEN, INC, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    long long result = ~0LL; \
+    void *ptr = buf; \
+    LOAD_pci_##SZ(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES1)); \
+    checkp(ptr, &buf[(1 * (INC)) % (LEN)]); \
+    LOAD_pci_##SZ(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES2)); \
+    checkp(ptr, &buf[(2 * (INC)) % (LEN)]); \
+    LOAD_pci_##SZ(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES3)); \
+    checkp(ptr, &buf[(3 * (INC)) % (LEN)]); \
+    LOAD_pci_##SZ(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES4)); \
+    checkp(ptr, &buf[(4 * (INC)) % (LEN)]); \
+}
+
+TEST_pci(loadalignb_pci, b, 2, 1,
+    0x01ffffffffffffffLL, 0x0201ffffffffffffLL,
+    0x010201ffffffffffLL, 0x02010201ffffffffLL)
+TEST_pci(loadalignh_pci, h, 4, 2,
+    0x0201ffffffffffffLL, 0x04030201ffffffffLL,
+    0x020104030201ffffLL, 0x0403020104030201LL)
+
+/*
+ ****************************************************************************
+ * _pcr addressing mode (addr ++ I:circ(modifier-reg))
+ */
+#define LOAD_pcr(SZ, RES, PTR, START, LEN, INC) \
+    __asm__( \
+        "r4 = %2\n\t" \
+        "m1 = r4\n\t" \
+        "cs1 = %3\n\t" \
+        "%0 = mem" #SZ "_fifo(%1++I:circ(m1))\n\t" \
+        : "+r"(RES), "+r"(PTR) \
+        : "r"((((INC) & 0x7f) << 17) | ((LEN) & 0x1ffff)), \
+          "r"(START) \
+        : "r4", "m1", "cs1")
+#define LOAD_pcr_b(RES, PTR, START, LEN, INC) \
+    LOAD_pcr(b, RES, PTR, START, LEN, INC)
+#define LOAD_pcr_h(RES, PTR, START, LEN, INC) \
+    LOAD_pcr(h, RES, PTR, START, LEN, INC)
+
+#define TEST_pcr(NAME, SZ, SIZE, LEN, INC, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    long long result = ~0LL; \
+    void *ptr = buf; \
+    LOAD_pcr_##SZ(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES1)); \
+    checkp(ptr, &buf[(1 * (INC) * (SIZE)) % (LEN)]); \
+    LOAD_pcr_##SZ(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES2)); \
+    checkp(ptr, &buf[(2 * (INC) * (SIZE)) % (LEN)]); \
+    LOAD_pcr_##SZ(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES3)); \
+    checkp(ptr, &buf[(3 * (INC) * (SIZE)) % (LEN)]); \
+    LOAD_pcr_##SZ(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES4)); \
+    checkp(ptr, &buf[(4 * (INC) * (SIZE)) % (LEN)]); \
+}
+
+TEST_pcr(loadalignb_pcr, b, 1, 2, 1,
+    0x01ffffffffffffffLL, 0x0201ffffffffffffLL,
+    0x010201ffffffffffLL, 0x02010201ffffffffLL)
+TEST_pcr(loadalignh_pcr, h, 2, 4, 1,
+    0x0201ffffffffffffLL, 0x04030201ffffffffLL,
+    0x020104030201ffffLL, 0x0403020104030201LL)
+
+int main()
+{
+    init_buf();
+
+    test_loadalignb_io();
+    test_loadalignh_io();
+
+    test_loadalignb_ur();
+    test_loadalignh_ur();
+
+    test_loadalignb_ap();
+    test_loadalignh_ap();
+
+    test_loadalignb_pr();
+    test_loadalignh_pr();
+
+    test_loadalignb_pbr();
+    test_loadalignh_pbr();
+
+    test_loadalignb_pi();
+    test_loadalignh_pi();
+
+    test_loadalignb_pci();
+    test_loadalignh_pci();
+
+    test_loadalignb_pcr();
+    test_loadalignh_pcr();
+
+    puts(err ? "FAIL" : "PASS");
+    return err ? 1 : 0;
+}
diff --git a/tests/tcg/hexagon/load_unpack.c b/tests/tcg/hexagon/load_unpack.c
new file mode 100644
index 0000000..3575a37
--- /dev/null
+++ b/tests/tcg/hexagon/load_unpack.c
@@ -0,0 +1,474 @@
+/*
+ *  Copyright(c) 2019-2021 Qualcomm Innovation Center, Inc. All Rights Reserved.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+/*
+ * Test load unpack instructions
+ *
+ * Example
+ *     r0 = memubh(r1+#0)
+ * loads a half word from memory and zero-extends the 2 bytes to form a word
+ *
+ * For each addressing mode, there are 4 tests
+ *     bzw2          unsigned     2 elements
+ *     bsw2          signed       2 elements
+ *     bzw4          unsigned     4 elements
+ *     bsw4          signed       4 elements
+ * There are 8 addressing modes, for a total of 32 instructions to test
+ */
+
+#include <stdio.h>
+#include <string.h>
+
+int err;
+
+char buf[16] __attribute__((aligned(1 << 16)));
+
+void init_buf(void)
+{
+    int i;
+    for (i = 0; i < 16; i++) {
+        int sign = i % 2 == 0 ? 0x80 : 0;
+        buf[i] = sign | (i + 1);
+    }
+}
+
+void __check(int line, long long result, long long expect)
+{
+    if (result != expect) {
+        printf("ERROR at line %d: 0x%08llx != 0x%08llx\n",
+               line, result, expect);
+        err++;
+    }
+}
+
+#define check(RES, EXP) __check(__LINE__, RES, EXP)
+
+void __checkp(int line, void *p, void *expect)
+{
+    if (p != expect) {
+        printf("ERROR at line %d: 0x%p != 0x%p\n", line, p, expect);
+        err++;
+    }
+}
+
+#define checkp(RES, EXP) __checkp(__LINE__, RES, EXP)
+
+/*
+ ****************************************************************************
+ * _io addressing mode (addr + offset)
+ */
+#define BxW_LOAD_io(SZ, RES, ADDR, OFF) \
+    __asm__( \
+        "%0 = mem" #SZ "(%1+#" #OFF ")\n\t" \
+        : "=r"(RES) \
+        : "r"(ADDR))
+#define BxW_LOAD_io_Z(RES, ADDR, OFF) \
+    BxW_LOAD_io(ubh, RES, ADDR, OFF)
+#define BxW_LOAD_io_S(RES, ADDR, OFF) \
+    BxW_LOAD_io(bh, RES, ADDR, OFF)
+
+#define TEST_io(NAME, TYPE, SIGN, SIZE, EXT, EXP1, EXP2, EXP3, EXP4) \
+void test_##NAME(void) \
+{ \
+    TYPE result; \
+    init_buf(); \
+    BxW_LOAD_io_##SIGN(result, buf, 0 * (SIZE)); \
+    check(result, (EXP1) | (EXT)); \
+    BxW_LOAD_io_##SIGN(result, buf, 1 * (SIZE)); \
+    check(result, (EXP2) | (EXT)); \
+    BxW_LOAD_io_##SIGN(result, buf, 2 * (SIZE)); \
+    check(result, (EXP3) | (EXT)); \
+    BxW_LOAD_io_##SIGN(result, buf, 3 * (SIZE)); \
+    check(result, (EXP4) | (EXT)); \
+}
+
+
+TEST_io(loadbzw2_io, int, Z, 2, 0x00000000,
+        0x00020081, 0x00040083, 0x00060085, 0x00080087)
+TEST_io(loadbsw2_io, int, S, 2, 0x0000ff00,
+        0x00020081, 0x00040083, 0x00060085, 0x00080087)
+TEST_io(loadbzw4_io, long long, Z,  4, 0x0000000000000000LL,
+        0x0004008300020081LL, 0x0008008700060085LL,
+        0x000c008b000a0089LL, 0x0010008f000e008dLL)
+TEST_io(loadbsw4_io, long long, S,  4, 0x0000ff000000ff00LL,
+        0x0004008300020081LL, 0x0008008700060085LL,
+        0x000c008b000a0089LL, 0x0010008f000e008dLL)
+
+/*
+ ****************************************************************************
+ * _ur addressing mode (index << offset + base)
+ */
+#define BxW_LOAD_ur(SZ, RES, SHIFT, IDX) \
+    __asm__( \
+        "%0 = mem" #SZ "(%1<<#" #SHIFT " + ##buf)\n\t" \
+        : "=r"(RES) \
+        : "r"(IDX))
+#define BxW_LOAD_ur_Z(RES, SHIFT, IDX) \
+    BxW_LOAD_ur(ubh, RES, SHIFT, IDX)
+#define BxW_LOAD_ur_S(RES, SHIFT, IDX) \
+    BxW_LOAD_ur(bh, RES, SHIFT, IDX)
+
+#define TEST_ur(NAME, TYPE, SIGN, SHIFT, EXT, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    TYPE result; \
+    init_buf(); \
+    BxW_LOAD_ur_##SIGN(result, (SHIFT), 0); \
+    check(result, (RES1) | (EXT)); \
+    BxW_LOAD_ur_##SIGN(result, (SHIFT), 1); \
+    check(result, (RES2) | (EXT)); \
+    BxW_LOAD_ur_##SIGN(result, (SHIFT), 2); \
+    check(result, (RES3) | (EXT)); \
+    BxW_LOAD_ur_##SIGN(result, (SHIFT), 3); \
+    check(result, (RES4) | (EXT)); \
+} \
+
+TEST_ur(loadbzw2_ur, int, Z, 1, 0x00000000,
+        0x00020081, 0x00040083, 0x00060085, 0x00080087)
+TEST_ur(loadbsw2_ur, int, S, 1, 0x0000ff00,
+        0x00020081, 0x00040083, 0x00060085, 0x00080087)
+TEST_ur(loadbzw4_ur, long long, Z, 2, 0x0000000000000000LL,
+        0x0004008300020081LL, 0x0008008700060085LL,
+        0x000c008b000a0089LL, 0x0010008f000e008dLL)
+TEST_ur(loadbsw4_ur, long long, S, 2, 0x0000ff000000ff00LL,
+        0x0004008300020081LL, 0x0008008700060085LL,
+        0x000c008b000a0089LL, 0x0010008f000e008dLL)
+
+/*
+ ****************************************************************************
+ * _ap addressing mode (addr = base)
+ */
+#define BxW_LOAD_ap(SZ, RES, PTR, ADDR) \
+    __asm__( \
+        "%0 = mem" #SZ "(%1 = ##" #ADDR ")\n\t" \
+        : "=r"(RES), "=r"(PTR))
+#define BxW_LOAD_ap_Z(RES, PTR, ADDR) \
+    BxW_LOAD_ap(ubh, RES, PTR, ADDR)
+#define BxW_LOAD_ap_S(RES, PTR, ADDR) \
+    BxW_LOAD_ap(bh, RES, PTR, ADDR)
+
+#define TEST_ap(NAME, TYPE, SIGN, SIZE, EXT, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    TYPE result; \
+    void *ptr; \
+    init_buf(); \
+    BxW_LOAD_ap_##SIGN(result, ptr, (buf + 0 * (SIZE))); \
+    check(result, (RES1) | (EXT)); \
+    checkp(ptr, &buf[0 * (SIZE)]); \
+    BxW_LOAD_ap_##SIGN(result, ptr, (buf + 1 * (SIZE))); \
+    check(result, (RES2) | (EXT)); \
+    checkp(ptr, &buf[1 * (SIZE)]); \
+    BxW_LOAD_ap_##SIGN(result, ptr, (buf + 2 * (SIZE))); \
+    check(result, (RES3) | (EXT)); \
+    checkp(ptr, &buf[2 * (SIZE)]); \
+    BxW_LOAD_ap_##SIGN(result, ptr, (buf + 3 * (SIZE))); \
+    check(result, (RES4) | (EXT)); \
+    checkp(ptr, &buf[3 * (SIZE)]); \
+}
+
+TEST_ap(loadbzw2_ap, int, Z, 2, 0x00000000,
+        0x00020081, 0x00040083, 0x00060085, 0x00080087)
+TEST_ap(loadbsw2_ap, int, S, 2, 0x0000ff00,
+        0x00020081, 0x00040083, 0x00060085, 0x00080087)
+TEST_ap(loadbzw4_ap, long long, Z, 4, 0x0000000000000000LL,
+        0x0004008300020081LL, 0x0008008700060085LL,
+        0x000c008b000a0089LL, 0x0010008f000e008dLL)
+TEST_ap(loadbsw4_ap, long long, S, 4, 0x0000ff000000ff00LL,
+        0x0004008300020081LL, 0x0008008700060085LL,
+        0x000c008b000a0089LL, 0x0010008f000e008dLL)
+
+/*
+ ****************************************************************************
+ * _rp addressing mode (addr ++ modifer-reg)
+ */
+#define BxW_LOAD_pr(SZ, RES, PTR, INC) \
+    __asm__( \
+        "m0 = %2\n\t" \
+        "%0 = mem" #SZ "(%1++m0)\n\t" \
+        : "=r"(RES), "+r"(PTR) \
+        : "r"(INC) \
+        : "m0")
+#define BxW_LOAD_pr_Z(RES, PTR, INC) \
+    BxW_LOAD_pr(ubh, RES, PTR, INC)
+#define BxW_LOAD_pr_S(RES, PTR, INC) \
+    BxW_LOAD_pr(bh, RES, PTR, INC)
+
+#define TEST_pr(NAME, TYPE, SIGN, SIZE, EXT, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    TYPE result; \
+    void *ptr = buf; \
+    init_buf(); \
+    BxW_LOAD_pr_##SIGN(result, ptr, (SIZE)); \
+    check(result, (RES1) | (EXT)); \
+    checkp(ptr, &buf[1 * (SIZE)]); \
+    BxW_LOAD_pr_##SIGN(result, ptr, (SIZE)); \
+    check(result, (RES2) | (EXT)); \
+    checkp(ptr, &buf[2 * (SIZE)]); \
+    BxW_LOAD_pr_##SIGN(result, ptr, (SIZE)); \
+    check(result, (RES3) | (EXT)); \
+    checkp(ptr, &buf[3 * (SIZE)]); \
+    BxW_LOAD_pr_##SIGN(result, ptr, (SIZE)); \
+    check(result, (RES4) | (EXT)); \
+    checkp(ptr, &buf[4 * (SIZE)]); \
+}
+
+TEST_pr(loadbzw2_pr, int, Z, 2, 0x00000000,
+    0x00020081, 0x0040083, 0x00060085, 0x00080087)
+TEST_pr(loadbsw2_pr, int, S, 2, 0x0000ff00,
+    0x00020081, 0x0040083, 0x00060085, 0x00080087)
+TEST_pr(loadbzw4_pr, long long, Z, 4, 0x0000000000000000LL,
+    0x0004008300020081LL, 0x0008008700060085LL,
+    0x000c008b000a0089LL, 0x0010008f000e008dLL)
+TEST_pr(loadbsw4_pr, long long, S, 4, 0x0000ff000000ff00LL,
+    0x0004008300020081LL, 0x0008008700060085LL,
+    0x000c008b000a0089LL, 0x0010008f000e008dLL)
+
+/*
+ ****************************************************************************
+ * _pbr addressing mode (addr ++ modifer-reg:brev)
+ */
+#define BxW_LOAD_pbr(SZ, RES, PTR) \
+    __asm__( \
+        "r4 = #(1 << (16 - 3))\n\t" \
+        "m0 = r4\n\t" \
+        "%0 = mem" #SZ "(%1++m0:brev)\n\t" \
+        : "=r"(RES), "+r"(PTR) \
+        : \
+        : "r4", "m0")
+#define BxW_LOAD_pbr_Z(RES, PTR) \
+    BxW_LOAD_pbr(ubh, RES, PTR)
+#define BxW_LOAD_pbr_S(RES, PTR) \
+    BxW_LOAD_pbr(bh, RES, PTR)
+
+#define TEST_pbr(NAME, TYPE, SIGN, EXT, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    TYPE result; \
+    void *ptr = buf; \
+    init_buf(); \
+    BxW_LOAD_pbr_##SIGN(result, ptr); \
+    check(result, (RES1) | (EXT)); \
+    BxW_LOAD_pbr_##SIGN(result, ptr); \
+    check(result, (RES2) | (EXT)); \
+    BxW_LOAD_pbr_##SIGN(result, ptr); \
+    check(result, (RES3) | (EXT)); \
+    BxW_LOAD_pbr_##SIGN(result, ptr); \
+    check(result, (RES4) | (EXT)); \
+}
+
+TEST_pbr(loadbzw2_pbr, int, Z, 0x00000000,
+    0x00020081, 0x00060085, 0x00040083, 0x00080087)
+TEST_pbr(loadbsw2_pbr, int, S, 0x0000ff00,
+    0x00020081, 0x00060085, 0x00040083, 0x00080087)
+TEST_pbr(loadbzw4_pbr, long long, Z, 0x0000000000000000LL,
+    0x0004008300020081LL, 0x0008008700060085LL,
+    0x0006008500040083LL, 0x000a008900080087LL)
+TEST_pbr(loadbsw4_pbr, long long, S, 0x0000ff000000ff00LL,
+    0x0004008300020081LL, 0x0008008700060085LL,
+    0x0006008500040083LL, 0x000a008900080087LL)
+
+/*
+ ****************************************************************************
+ * _pi addressing mode (addr ++ inc)
+ */
+#define BxW_LOAD_pi(SZ, RES, PTR, INC) \
+    __asm__( \
+        "%0 = mem" #SZ "(%1++#" #INC ")\n\t" \
+        : "=r"(RES), "+r"(PTR))
+#define BxW_LOAD_pi_Z(RES, PTR, INC) \
+    BxW_LOAD_pi(ubh, RES, PTR, INC)
+#define BxW_LOAD_pi_S(RES, PTR, INC) \
+    BxW_LOAD_pi(bh, RES, PTR, INC)
+
+#define TEST_pi(NAME, TYPE, SIGN, INC, EXT, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    TYPE result; \
+    void *ptr = buf; \
+    init_buf(); \
+    BxW_LOAD_pi_##SIGN(result, ptr, (INC)); \
+    check(result, (RES1) | (EXT)); \
+    checkp(ptr, &buf[1 * (INC)]); \
+    BxW_LOAD_pi_##SIGN(result, ptr, (INC)); \
+    check(result, (RES2) | (EXT)); \
+    checkp(ptr, &buf[2 * (INC)]); \
+    BxW_LOAD_pi_##SIGN(result, ptr, (INC)); \
+    check(result, (RES3) | (EXT)); \
+    checkp(ptr, &buf[3 * (INC)]); \
+    BxW_LOAD_pi_##SIGN(result, ptr, (INC)); \
+    check(result, (RES4) | (EXT)); \
+    checkp(ptr, &buf[4 * (INC)]); \
+}
+
+TEST_pi(loadbzw2_pi, int, Z, 2, 0x00000000,
+    0x00020081, 0x00040083, 0x00060085, 0x00080087)
+TEST_pi(loadbsw2_pi, int, S, 2, 0x0000ff00,
+    0x00020081, 0x00040083, 0x00060085, 0x00080087)
+TEST_pi(loadbzw4_pi, long long, Z, 4, 0x0000000000000000LL,
+    0x0004008300020081LL, 0x0008008700060085LL,
+    0x000c008b000a0089LL, 0x0010008f000e008dLL)
+TEST_pi(loadbsw4_pi, long long, S, 4, 0x0000ff000000ff00LL,
+    0x0004008300020081LL, 0x0008008700060085LL,
+    0x000c008b000a0089LL, 0x0010008f000e008dLL)
+
+/*
+ ****************************************************************************
+ * _pci addressing mode (addr ++ inc:circ)
+ */
+#define BxW_LOAD_pci(SZ, RES, PTR, START, LEN, INC) \
+    __asm__( \
+        "r4 = %3\n\t" \
+        "m0 = r4\n\t" \
+        "cs0 = %2\n\t" \
+        "%0 = mem" #SZ "(%1++#" #INC ":circ(m0))\n\t" \
+        : "=r"(RES), "+r"(PTR) \
+        : "r"(START), "r"(LEN) \
+        : "r4", "m0", "cs0")
+#define BxW_LOAD_pci_Z(RES, PTR, START, LEN, INC) \
+    BxW_LOAD_pci(ubh, RES, PTR, START, LEN, INC)
+#define BxW_LOAD_pci_S(RES, PTR, START, LEN, INC) \
+    BxW_LOAD_pci(bh, RES, PTR, START, LEN, INC)
+
+#define TEST_pci(NAME, TYPE, SIGN, LEN, INC, EXT, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    TYPE result; \
+    void *ptr = buf; \
+    init_buf(); \
+    BxW_LOAD_pci_##SIGN(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES1) | (EXT)); \
+    checkp(ptr, &buf[(1 * (INC)) % (LEN)]); \
+    BxW_LOAD_pci_##SIGN(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES2) | (EXT)); \
+    checkp(ptr, &buf[(2 * (INC)) % (LEN)]); \
+    BxW_LOAD_pci_##SIGN(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES3) | (EXT)); \
+    checkp(ptr, &buf[(3 * (INC)) % (LEN)]); \
+    BxW_LOAD_pci_##SIGN(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES4) | (EXT)); \
+    checkp(ptr, &buf[(4 * (INC)) % (LEN)]); \
+}
+
+TEST_pci(loadbzw2_pci, int, Z, 6, 2, 0x00000000,
+    0x00020081, 0x00040083, 0x00060085, 0x00020081)
+TEST_pci(loadbsw2_pci, int, S, 6, 2, 0x0000ff00,
+    0x00020081, 0x00040083, 0x00060085, 0x00020081)
+TEST_pci(loadbzw4_pci, long long, Z, 8, 4, 0x0000000000000000LL,
+    0x0004008300020081LL, 0x0008008700060085LL,
+    0x0004008300020081LL, 0x0008008700060085LL)
+TEST_pci(loadbsw4_pci, long long, S, 8, 4, 0x0000ff000000ff00LL,
+    0x0004008300020081LL, 0x0008008700060085LL,
+    0x0004008300020081LL, 0x0008008700060085LL)
+
+/*
+ ****************************************************************************
+ * _pcr addressing mode (addr ++ I:circ(modifier-reg))
+ */
+#define BxW_LOAD_pcr(SZ, RES, PTR, START, LEN, INC) \
+    __asm__( \
+        "r4 = %2\n\t" \
+        "m1 = r4\n\t" \
+        "cs1 = %3\n\t" \
+        "%0 = mem" #SZ "(%1++I:circ(m1))\n\t" \
+        : "=r"(RES), "+r"(PTR) \
+        : "r"((((INC) & 0x7f) << 17) | ((LEN) & 0x1ffff)), \
+          "r"(START) \
+        : "r4", "m1", "cs1")
+#define BxW_LOAD_pcr_Z(RES, PTR, START, LEN, INC) \
+    BxW_LOAD_pcr(ubh, RES, PTR, START, LEN, INC)
+#define BxW_LOAD_pcr_S(RES, PTR, START, LEN, INC) \
+    BxW_LOAD_pcr(bh, RES, PTR, START, LEN, INC)
+
+#define TEST_pcr(NAME, TYPE, SIGN, SIZE, LEN, INC, \
+                 EXT, RES1, RES2, RES3, RES4) \
+void test_##NAME(void) \
+{ \
+    TYPE result; \
+    void *ptr = buf; \
+    init_buf(); \
+    BxW_LOAD_pcr_##SIGN(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES1) | (EXT)); \
+    checkp(ptr, &buf[(1 * (INC) * (SIZE)) % (LEN)]); \
+    BxW_LOAD_pcr_##SIGN(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES2) | (EXT)); \
+    checkp(ptr, &buf[(2 * (INC) * (SIZE)) % (LEN)]); \
+    BxW_LOAD_pcr_##SIGN(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES3) | (EXT)); \
+    checkp(ptr, &buf[(3 * (INC) * (SIZE)) % (LEN)]); \
+    BxW_LOAD_pcr_##SIGN(result, ptr, buf, (LEN), (INC)); \
+    check(result, (RES4) | (EXT)); \
+    checkp(ptr, &buf[(4 * (INC) * (SIZE)) % (LEN)]); \
+}
+
+TEST_pcr(loadbzw2_pcr, int, Z, 2, 8, 2, 0x00000000,
+    0x00020081, 0x00060085, 0x00020081, 0x00060085)
+TEST_pcr(loadbsw2_pcr, int, S, 2, 8, 2, 0x0000ff00,
+    0x00020081, 0x00060085, 0x00020081, 0x00060085)
+TEST_pcr(loadbzw4_pcr, long long, Z, 4, 8, 1, 0x0000000000000000LL,
+    0x0004008300020081LL, 0x0008008700060085LL,
+    0x0004008300020081LL, 0x0008008700060085LL)
+TEST_pcr(loadbsw4_pcr, long long, S, 4, 8, 1, 0x0000ff000000ff00LL,
+    0x0004008300020081LL, 0x0008008700060085LL,
+    0x0004008300020081LL, 0x0008008700060085LL)
+
+int main()
+{
+    test_loadbzw2_io();
+    test_loadbsw2_io();
+    test_loadbzw4_io();
+    test_loadbsw4_io();
+
+    test_loadbzw2_ur();
+    test_loadbsw2_ur();
+    test_loadbzw4_ur();
+    test_loadbsw4_ur();
+
+    test_loadbzw2_ap();
+    test_loadbsw2_ap();
+    test_loadbzw4_ap();
+    test_loadbsw4_ap();
+
+    test_loadbzw2_pr();
+    test_loadbsw2_pr();
+    test_loadbzw4_pr();
+    test_loadbsw4_pr();
+
+    test_loadbzw2_pbr();
+    test_loadbsw2_pbr();
+    test_loadbzw4_pbr();
+    test_loadbsw4_pbr();
+
+    test_loadbzw2_pi();
+    test_loadbsw2_pi();
+    test_loadbzw4_pi();
+    test_loadbsw4_pi();
+
+    test_loadbzw2_pci();
+    test_loadbsw2_pci();
+    test_loadbzw4_pci();
+    test_loadbsw4_pci();
+
+    test_loadbzw2_pcr();
+    test_loadbsw2_pcr();
+    test_loadbzw4_pcr();
+    test_loadbsw4_pcr();
+
+    puts(err ? "FAIL" : "PASS");
+    return err ? 1 : 0;
+}
diff --git a/tests/tcg/hexagon/misc.c b/tests/tcg/hexagon/misc.c
index 458759f..17c3919 100644
--- a/tests/tcg/hexagon/misc.c
+++ b/tests/tcg/hexagon/misc.c
@@ -231,6 +231,14 @@ static void check(int val, int expect)
     }
 }
 
+static void check64(long long val, long long expect)
+{
+    if (val != expect) {
+        printf("ERROR: 0x%016llx != 0x%016llx\n", val, expect);
+        err++;
+    }
+}
+
 uint32_t init[10] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };
 uint32_t array[10];
 
@@ -264,8 +272,36 @@ static long long creg_pair(int x, int y)
     return retval;
 }
 
+static long long decbin(long long x, long long y, int *pred)
+{
+    long long retval;
+    asm ("%0 = decbin(%2, %3)\n\t"
+         "%1 = p0\n\t"
+         : "=r"(retval), "=r"(*pred)
+         : "r"(x), "r"(y));
+    return retval;
+}
+
+/* Check that predicates are auto-and'ed in a packet */
+static int auto_and(void)
+{
+    int retval;
+    asm ("r5 = #1\n\t"
+         "{\n\t"
+         "    p0 = cmp.eq(r1, #1)\n\t"
+         "    p0 = cmp.eq(r1, #2)\n\t"
+         "}\n\t"
+         "%0 = p0\n\t"
+         : "=r"(retval)
+         :
+         : "r5", "p0");
+    return retval;
+}
+
 int main()
 {
+    long long res64;
+    int pred;
 
     memcpy(array, init, sizeof(array));
     S4_storerhnew_rr(array, 4, 0xffff);
@@ -375,6 +411,17 @@ int main()
     res = test_clrtnew(2, 7);
     check(res, 7);
 
+    res64 = decbin(0xf0f1f2f3f4f5f6f7LL, 0x7f6f5f4f3f2f1f0fLL, &pred);
+    check64(res64, 0x357980003700010cLL);
+    check(pred, 0);
+
+    res64 = decbin(0xfLL, 0x1bLL, &pred);
+    check64(res64, 0x78000100LL);
+    check(pred, 1);
+
+    res = auto_and();
+    check(res, 0);
+
     puts(err ? "FAIL" : "PASS");
     return err;
 }
diff --git a/tests/tcg/hexagon/multi_result.c b/tests/tcg/hexagon/multi_result.c
new file mode 100644
index 0000000..52997b3
--- /dev/null
+++ b/tests/tcg/hexagon/multi_result.c
@@ -0,0 +1,282 @@
+/*
+ *  Copyright(c) 2019-2021 Qualcomm Innovation Center, Inc. All Rights Reserved.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <stdio.h>
+
+static int sfrecipa(int Rs, int Rt, int *pred_result)
+{
+  int result;
+  int predval;
+
+  asm volatile("%0,p0 = sfrecipa(%2, %3)\n\t"
+               "%1 = p0\n\t"
+               : "+r"(result), "=r"(predval)
+               : "r"(Rs), "r"(Rt)
+               : "p0");
+  *pred_result = predval;
+  return result;
+}
+
+static int sfinvsqrta(int Rs, int *pred_result)
+{
+  int result;
+  int predval;
+
+  asm volatile("%0,p0 = sfinvsqrta(%2)\n\t"
+               "%1 = p0\n\t"
+               : "+r"(result), "=r"(predval)
+               : "r"(Rs)
+               : "p0");
+  *pred_result = predval;
+  return result;
+}
+
+static long long vacsh(long long Rxx, long long Rss, long long Rtt,
+                       int *pred_result, int *ovf_result)
+{
+  long long result = Rxx;
+  int predval;
+  int usr;
+
+  /*
+   * This instruction can set bit 0 (OVF/overflow) in usr
+   * Clear the bit first, then return that bit to the caller
+   */
+  asm volatile("r2 = usr\n\t"
+               "r2 = clrbit(r2, #0)\n\t"        /* clear overflow bit */
+               "usr = r2\n\t"
+               "%0,p0 = vacsh(%3, %4)\n\t"
+               "%1 = p0\n\t"
+               "%2 = usr\n\t"
+               : "+r"(result), "=r"(predval), "=r"(usr)
+               : "r"(Rss), "r"(Rtt)
+               : "r2", "p0", "usr");
+  *pred_result = predval;
+  *ovf_result = (usr & 1);
+  return result;
+}
+
+static long long vminub(long long Rtt, long long Rss,
+                        int *pred_result)
+{
+  long long result;
+  int predval;
+
+  asm volatile("%0,p0 = vminub(%2, %3)\n\t"
+               "%1 = p0\n\t"
+               : "=r"(result), "=r"(predval)
+               : "r"(Rtt), "r"(Rss)
+               : "p0");
+  *pred_result = predval;
+  return result;
+}
+
+static long long add_carry(long long Rss, long long Rtt,
+                           int pred_in, int *pred_result)
+{
+  long long result;
+  int predval = pred_in;
+
+  asm volatile("p0 = %1\n\t"
+               "%0 = add(%2, %3, p0):carry\n\t"
+               "%1 = p0\n\t"
+               : "=r"(result), "+r"(predval)
+               : "r"(Rss), "r"(Rtt)
+               : "p0");
+  *pred_result = predval;
+  return result;
+}
+
+static long long sub_carry(long long Rss, long long Rtt,
+                           int pred_in, int *pred_result)
+{
+  long long result;
+  int predval = pred_in;
+
+  asm volatile("p0 = !cmp.eq(%1, #0)\n\t"
+               "%0 = sub(%2, %3, p0):carry\n\t"
+               "%1 = p0\n\t"
+               : "=r"(result), "+r"(predval)
+               : "r"(Rss), "r"(Rtt)
+               : "p0");
+  *pred_result = predval;
+  return result;
+}
+
+int err;
+
+static void check_ll(long long val, long long expect)
+{
+    if (val != expect) {
+        printf("ERROR: 0x%016llx != 0x%016llx\n", val, expect);
+        err++;
+    }
+}
+
+static void check(int val, int expect)
+{
+    if (val != expect) {
+        printf("ERROR: 0x%08x != 0x%08x\n", val, expect);
+        err++;
+    }
+}
+
+static void check_p(int val, int expect)
+{
+    if (val != expect) {
+        printf("ERROR: 0x%02x != 0x%02x\n", val, expect);
+        err++;
+    }
+}
+
+static void test_sfrecipa()
+{
+    int res;
+    int pred_result;
+
+    res = sfrecipa(0x04030201, 0x05060708, &pred_result);
+    check(res, 0x59f38001);
+    check_p(pred_result, 0x00);
+}
+
+static void test_sfinvsqrta()
+{
+    int res;
+    int pred_result;
+
+    res = sfinvsqrta(0x04030201, &pred_result);
+    check(res, 0x4d330000);
+    check_p(pred_result, 0xe0);
+
+    res = sfinvsqrta(0x0, &pred_result);
+    check(res, 0x3f800000);
+    check_p(pred_result, 0x0);
+}
+
+static void test_vacsh()
+{
+    long long res64;
+    int pred_result;
+    int ovf_result;
+
+    res64 = vacsh(0x0004000300020001LL,
+                  0x0001000200030004LL,
+                  0x0000000000000000LL, &pred_result, &ovf_result);
+    check_ll(res64, 0x0004000300030004LL);
+    check_p(pred_result, 0xf0);
+    check(ovf_result, 0);
+
+    res64 = vacsh(0x0004000300020001LL,
+                  0x0001000200030004LL,
+                  0x000affff000d0000LL, &pred_result, &ovf_result);
+    check_ll(res64, 0x000e0003000f0004LL);
+    check_p(pred_result, 0xcc);
+    check(ovf_result, 0);
+
+    res64 = vacsh(0x00047fff00020001LL,
+                  0x00017fff00030004LL,
+                  0x000a0fff000d0000LL, &pred_result, &ovf_result);
+    check_ll(res64, 0x000e7fff000f0004LL);
+    check_p(pred_result, 0xfc);
+    check(ovf_result, 1);
+
+    res64 = vacsh(0x0004000300020001LL,
+                  0x0001000200030009LL,
+                  0x000affff000d0001LL, &pred_result, &ovf_result);
+    check_ll(res64, 0x000e0003000f0008LL);
+    check_p(pred_result, 0xcc);
+    check(ovf_result, 0);
+}
+
+static void test_vminub()
+{
+    long long res64;
+    int pred_result;
+
+    res64 = vminub(0x0807060504030201LL,
+                   0x0102030405060708LL,
+                   &pred_result);
+    check_ll(res64, 0x0102030404030201LL);
+    check_p(pred_result, 0xf0);
+
+    res64 = vminub(0x0802060405030701LL,
+                   0x0107030504060208LL,
+                   &pred_result);
+    check_ll(res64, 0x0102030404030201LL);
+    check_p(pred_result, 0xaa);
+}
+
+static void test_add_carry()
+{
+    long long res64;
+    int pred_result;
+
+    res64 = add_carry(0x0000000000000000LL,
+                      0xffffffffffffffffLL,
+                      1, &pred_result);
+    check_ll(res64, 0x0000000000000000LL);
+    check_p(pred_result, 0xff);
+
+    res64 = add_carry(0x0000000100000000LL,
+                      0xffffffffffffffffLL,
+                      0, &pred_result);
+    check_ll(res64, 0x00000000ffffffffLL);
+    check_p(pred_result, 0xff);
+
+    res64 = add_carry(0x0000000100000000LL,
+                      0xffffffffffffffffLL,
+                      0, &pred_result);
+    check_ll(res64, 0x00000000ffffffffLL);
+    check_p(pred_result, 0xff);
+}
+
+static void test_sub_carry()
+{
+    long long res64;
+    int pred_result;
+
+    res64 = sub_carry(0x0000000000000000LL,
+                      0x0000000000000000LL,
+                      1, &pred_result);
+    check_ll(res64, 0x0000000000000000LL);
+    check_p(pred_result, 0xff);
+
+    res64 = sub_carry(0x0000000100000000LL,
+                      0x0000000000000000LL,
+                      0, &pred_result);
+    check_ll(res64, 0x00000000ffffffffLL);
+    check_p(pred_result, 0xff);
+
+    res64 = sub_carry(0x0000000100000000LL,
+                      0x0000000000000000LL,
+                      0, &pred_result);
+    check_ll(res64, 0x00000000ffffffffLL);
+    check_p(pred_result, 0xff);
+}
+
+int main()
+{
+    test_sfrecipa();
+    test_sfinvsqrta();
+    test_vacsh();
+    test_vminub();
+    test_add_carry();
+    test_sub_carry();
+
+    puts(err ? "FAIL" : "PASS");
+    return err;
+}