Merge remote-tracking branch 'remotes/amarkovic/tags/mips-queue-may-19-2019-v3' into staging

MIPS queue for May 19th, 2019 - v3

# gpg: Signature made Sun 26 May 2019 17:07:07 BST
# gpg:                using RSA key D4972A8967F75A65
# gpg: Good signature from "Aleksandar Markovic <amarkovic@wavecomp.com>" [unknown]
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 8526 FBF1 5DA3 811F 4A01  DD75 D497 2A89 67F7 5A65

* remotes/amarkovic/tags/mips-queue-may-19-2019-v3:
  BootLinuxSshTest: Test some userspace commands on Malta
  target/mips: realign comments to fix checkpatch warnings
  target/mips: add or remove space to fix checkpatch errors
  linux-user: fix __NR_semtimedop undeclared error
  mips: Decide to map PAGE_EXEC in map_address
  target/mips: Refactor and fix INSERT.<B|H|W|D> instructions
  target/mips: Refactor and fix COPY_U.<B|H|W> instructions
  target/mips: Refactor and fix COPY_S.<B|H|W|D> instructions
  target/mips: Fix MSA instructions ST.<B|H|W|D> on big endian host
  target/mips: Fix MSA instructions LD.<B|H|W|D> on big endian host
  target/mips: Make the results of MOD_<U|S>.<B|H|W|D> the same as on hardware
  target/mips: Make the results of DIV_<U|S>.<B|H|W|D> the same as on hardware

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

1 files changed, 348 insertions, 40 deletions

diff --git a/target/mips/op_helper.c b/target/mips/op_helper.c
index 6d86912958..39180275b5 100644
--- a/target/mips/op_helper.c
+++ b/target/mips/op_helper.c
@@ -4356,31 +4356,179 @@ FOP_CONDN_S(sne,  (float32_lt(fst1, fst0, &env->active_fpu.fp_status)
 #define MEMOP_IDX(DF)
 #endif
 
-#define MSA_LD_DF(DF, TYPE, LD_INSN, ...)                               \
-void helper_msa_ld_ ## TYPE(CPUMIPSState *env, uint32_t wd,             \
-                            target_ulong addr)                          \
-{                                                                       \
-    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);                          \
-    wr_t wx;                                                            \
-    int i;                                                              \
-    MEMOP_IDX(DF)                                                       \
-    for (i = 0; i < DF_ELEMENTS(DF); i++) {                             \
-        wx.TYPE[i] = LD_INSN(env, addr + (i << DF), ##__VA_ARGS__);     \
-    }                                                                   \
-    memcpy(pwd, &wx, sizeof(wr_t));                                     \
+void helper_msa_ld_b(CPUMIPSState *env, uint32_t wd,
+                     target_ulong addr)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    MEMOP_IDX(DF_BYTE)
+#if !defined(CONFIG_USER_ONLY)
+#if !defined(HOST_WORDS_BIGENDIAN)
+    pwd->b[0]  = helper_ret_ldub_mmu(env, addr + (0  << DF_BYTE), oi, GETPC());
+    pwd->b[1]  = helper_ret_ldub_mmu(env, addr + (1  << DF_BYTE), oi, GETPC());
+    pwd->b[2]  = helper_ret_ldub_mmu(env, addr + (2  << DF_BYTE), oi, GETPC());
+    pwd->b[3]  = helper_ret_ldub_mmu(env, addr + (3  << DF_BYTE), oi, GETPC());
+    pwd->b[4]  = helper_ret_ldub_mmu(env, addr + (4  << DF_BYTE), oi, GETPC());
+    pwd->b[5]  = helper_ret_ldub_mmu(env, addr + (5  << DF_BYTE), oi, GETPC());
+    pwd->b[6]  = helper_ret_ldub_mmu(env, addr + (6  << DF_BYTE), oi, GETPC());
+    pwd->b[7]  = helper_ret_ldub_mmu(env, addr + (7  << DF_BYTE), oi, GETPC());
+    pwd->b[8]  = helper_ret_ldub_mmu(env, addr + (8  << DF_BYTE), oi, GETPC());
+    pwd->b[9]  = helper_ret_ldub_mmu(env, addr + (9  << DF_BYTE), oi, GETPC());
+    pwd->b[10] = helper_ret_ldub_mmu(env, addr + (10 << DF_BYTE), oi, GETPC());
+    pwd->b[11] = helper_ret_ldub_mmu(env, addr + (11 << DF_BYTE), oi, GETPC());
+    pwd->b[12] = helper_ret_ldub_mmu(env, addr + (12 << DF_BYTE), oi, GETPC());
+    pwd->b[13] = helper_ret_ldub_mmu(env, addr + (13 << DF_BYTE), oi, GETPC());
+    pwd->b[14] = helper_ret_ldub_mmu(env, addr + (14 << DF_BYTE), oi, GETPC());
+    pwd->b[15] = helper_ret_ldub_mmu(env, addr + (15 << DF_BYTE), oi, GETPC());
+#else
+    pwd->b[0]  = helper_ret_ldub_mmu(env, addr + (7  << DF_BYTE), oi, GETPC());
+    pwd->b[1]  = helper_ret_ldub_mmu(env, addr + (6  << DF_BYTE), oi, GETPC());
+    pwd->b[2]  = helper_ret_ldub_mmu(env, addr + (5  << DF_BYTE), oi, GETPC());
+    pwd->b[3]  = helper_ret_ldub_mmu(env, addr + (4  << DF_BYTE), oi, GETPC());
+    pwd->b[4]  = helper_ret_ldub_mmu(env, addr + (3  << DF_BYTE), oi, GETPC());
+    pwd->b[5]  = helper_ret_ldub_mmu(env, addr + (2  << DF_BYTE), oi, GETPC());
+    pwd->b[6]  = helper_ret_ldub_mmu(env, addr + (1  << DF_BYTE), oi, GETPC());
+    pwd->b[7]  = helper_ret_ldub_mmu(env, addr + (0  << DF_BYTE), oi, GETPC());
+    pwd->b[8]  = helper_ret_ldub_mmu(env, addr + (15 << DF_BYTE), oi, GETPC());
+    pwd->b[9]  = helper_ret_ldub_mmu(env, addr + (14 << DF_BYTE), oi, GETPC());
+    pwd->b[10] = helper_ret_ldub_mmu(env, addr + (13 << DF_BYTE), oi, GETPC());
+    pwd->b[11] = helper_ret_ldub_mmu(env, addr + (12 << DF_BYTE), oi, GETPC());
+    pwd->b[12] = helper_ret_ldub_mmu(env, addr + (11 << DF_BYTE), oi, GETPC());
+    pwd->b[13] = helper_ret_ldub_mmu(env, addr + (10 << DF_BYTE), oi, GETPC());
+    pwd->b[14] = helper_ret_ldub_mmu(env, addr + (9  << DF_BYTE), oi, GETPC());
+    pwd->b[15] = helper_ret_ldub_mmu(env, addr + (8  << DF_BYTE), oi, GETPC());
+#endif
+#else
+#if !defined(HOST_WORDS_BIGENDIAN)
+    pwd->b[0]  = cpu_ldub_data(env, addr + (0  << DF_BYTE));
+    pwd->b[1]  = cpu_ldub_data(env, addr + (1  << DF_BYTE));
+    pwd->b[2]  = cpu_ldub_data(env, addr + (2  << DF_BYTE));
+    pwd->b[3]  = cpu_ldub_data(env, addr + (3  << DF_BYTE));
+    pwd->b[4]  = cpu_ldub_data(env, addr + (4  << DF_BYTE));
+    pwd->b[5]  = cpu_ldub_data(env, addr + (5  << DF_BYTE));
+    pwd->b[6]  = cpu_ldub_data(env, addr + (6  << DF_BYTE));
+    pwd->b[7]  = cpu_ldub_data(env, addr + (7  << DF_BYTE));
+    pwd->b[8]  = cpu_ldub_data(env, addr + (8  << DF_BYTE));
+    pwd->b[9]  = cpu_ldub_data(env, addr + (9  << DF_BYTE));
+    pwd->b[10] = cpu_ldub_data(env, addr + (10 << DF_BYTE));
+    pwd->b[11] = cpu_ldub_data(env, addr + (11 << DF_BYTE));
+    pwd->b[12] = cpu_ldub_data(env, addr + (12 << DF_BYTE));
+    pwd->b[13] = cpu_ldub_data(env, addr + (13 << DF_BYTE));
+    pwd->b[14] = cpu_ldub_data(env, addr + (14 << DF_BYTE));
+    pwd->b[15] = cpu_ldub_data(env, addr + (15 << DF_BYTE));
+#else
+    pwd->b[0]  = cpu_ldub_data(env, addr + (7  << DF_BYTE));
+    pwd->b[1]  = cpu_ldub_data(env, addr + (6  << DF_BYTE));
+    pwd->b[2]  = cpu_ldub_data(env, addr + (5  << DF_BYTE));
+    pwd->b[3]  = cpu_ldub_data(env, addr + (4  << DF_BYTE));
+    pwd->b[4]  = cpu_ldub_data(env, addr + (3  << DF_BYTE));
+    pwd->b[5]  = cpu_ldub_data(env, addr + (2  << DF_BYTE));
+    pwd->b[6]  = cpu_ldub_data(env, addr + (1  << DF_BYTE));
+    pwd->b[7]  = cpu_ldub_data(env, addr + (0  << DF_BYTE));
+    pwd->b[8]  = cpu_ldub_data(env, addr + (15 << DF_BYTE));
+    pwd->b[9]  = cpu_ldub_data(env, addr + (14 << DF_BYTE));
+    pwd->b[10] = cpu_ldub_data(env, addr + (13 << DF_BYTE));
+    pwd->b[11] = cpu_ldub_data(env, addr + (12 << DF_BYTE));
+    pwd->b[12] = cpu_ldub_data(env, addr + (11 << DF_BYTE));
+    pwd->b[13] = cpu_ldub_data(env, addr + (10 << DF_BYTE));
+    pwd->b[14] = cpu_ldub_data(env, addr + (9 << DF_BYTE));
+    pwd->b[15] = cpu_ldub_data(env, addr + (8 << DF_BYTE));
+#endif
+#endif
+}
+
+void helper_msa_ld_h(CPUMIPSState *env, uint32_t wd,
+                     target_ulong addr)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    MEMOP_IDX(DF_HALF)
+#if !defined(CONFIG_USER_ONLY)
+#if !defined(HOST_WORDS_BIGENDIAN)
+    pwd->h[0] = helper_ret_lduw_mmu(env, addr + (0 << DF_HALF), oi, GETPC());
+    pwd->h[1] = helper_ret_lduw_mmu(env, addr + (1 << DF_HALF), oi, GETPC());
+    pwd->h[2] = helper_ret_lduw_mmu(env, addr + (2 << DF_HALF), oi, GETPC());
+    pwd->h[3] = helper_ret_lduw_mmu(env, addr + (3 << DF_HALF), oi, GETPC());
+    pwd->h[4] = helper_ret_lduw_mmu(env, addr + (4 << DF_HALF), oi, GETPC());
+    pwd->h[5] = helper_ret_lduw_mmu(env, addr + (5 << DF_HALF), oi, GETPC());
+    pwd->h[6] = helper_ret_lduw_mmu(env, addr + (6 << DF_HALF), oi, GETPC());
+    pwd->h[7] = helper_ret_lduw_mmu(env, addr + (7 << DF_HALF), oi, GETPC());
+#else
+    pwd->h[0] = helper_ret_lduw_mmu(env, addr + (3 << DF_HALF), oi, GETPC());
+    pwd->h[1] = helper_ret_lduw_mmu(env, addr + (2 << DF_HALF), oi, GETPC());
+    pwd->h[2] = helper_ret_lduw_mmu(env, addr + (1 << DF_HALF), oi, GETPC());
+    pwd->h[3] = helper_ret_lduw_mmu(env, addr + (0 << DF_HALF), oi, GETPC());
+    pwd->h[4] = helper_ret_lduw_mmu(env, addr + (7 << DF_HALF), oi, GETPC());
+    pwd->h[5] = helper_ret_lduw_mmu(env, addr + (6 << DF_HALF), oi, GETPC());
+    pwd->h[6] = helper_ret_lduw_mmu(env, addr + (5 << DF_HALF), oi, GETPC());
+    pwd->h[7] = helper_ret_lduw_mmu(env, addr + (4 << DF_HALF), oi, GETPC());
+#endif
+#else
+#if !defined(HOST_WORDS_BIGENDIAN)
+    pwd->h[0] = cpu_lduw_data(env, addr + (0 << DF_HALF));
+    pwd->h[1] = cpu_lduw_data(env, addr + (1 << DF_HALF));
+    pwd->h[2] = cpu_lduw_data(env, addr + (2 << DF_HALF));
+    pwd->h[3] = cpu_lduw_data(env, addr + (3 << DF_HALF));
+    pwd->h[4] = cpu_lduw_data(env, addr + (4 << DF_HALF));
+    pwd->h[5] = cpu_lduw_data(env, addr + (5 << DF_HALF));
+    pwd->h[6] = cpu_lduw_data(env, addr + (6 << DF_HALF));
+    pwd->h[7] = cpu_lduw_data(env, addr + (7 << DF_HALF));
+#else
+    pwd->h[0] = cpu_lduw_data(env, addr + (3 << DF_HALF));
+    pwd->h[1] = cpu_lduw_data(env, addr + (2 << DF_HALF));
+    pwd->h[2] = cpu_lduw_data(env, addr + (1 << DF_HALF));
+    pwd->h[3] = cpu_lduw_data(env, addr + (0 << DF_HALF));
+    pwd->h[4] = cpu_lduw_data(env, addr + (7 << DF_HALF));
+    pwd->h[5] = cpu_lduw_data(env, addr + (6 << DF_HALF));
+    pwd->h[6] = cpu_lduw_data(env, addr + (5 << DF_HALF));
+    pwd->h[7] = cpu_lduw_data(env, addr + (4 << DF_HALF));
+#endif
+#endif
 }
 
+void helper_msa_ld_w(CPUMIPSState *env, uint32_t wd,
+                     target_ulong addr)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    MEMOP_IDX(DF_WORD)
 #if !defined(CONFIG_USER_ONLY)
-MSA_LD_DF(DF_BYTE,   b, helper_ret_ldub_mmu, oi, GETPC())
-MSA_LD_DF(DF_HALF,   h, helper_ret_lduw_mmu, oi, GETPC())
-MSA_LD_DF(DF_WORD,   w, helper_ret_ldul_mmu, oi, GETPC())
-MSA_LD_DF(DF_DOUBLE, d, helper_ret_ldq_mmu,  oi, GETPC())
+#if !defined(HOST_WORDS_BIGENDIAN)
+    pwd->w[0] = helper_ret_ldul_mmu(env, addr + (0 << DF_WORD), oi, GETPC());
+    pwd->w[1] = helper_ret_ldul_mmu(env, addr + (1 << DF_WORD), oi, GETPC());
+    pwd->w[2] = helper_ret_ldul_mmu(env, addr + (2 << DF_WORD), oi, GETPC());
+    pwd->w[3] = helper_ret_ldul_mmu(env, addr + (3 << DF_WORD), oi, GETPC());
 #else
-MSA_LD_DF(DF_BYTE,   b, cpu_ldub_data)
-MSA_LD_DF(DF_HALF,   h, cpu_lduw_data)
-MSA_LD_DF(DF_WORD,   w, cpu_ldl_data)
-MSA_LD_DF(DF_DOUBLE, d, cpu_ldq_data)
+    pwd->w[0] = helper_ret_ldul_mmu(env, addr + (1 << DF_WORD), oi, GETPC());
+    pwd->w[1] = helper_ret_ldul_mmu(env, addr + (0 << DF_WORD), oi, GETPC());
+    pwd->w[2] = helper_ret_ldul_mmu(env, addr + (3 << DF_WORD), oi, GETPC());
+    pwd->w[3] = helper_ret_ldul_mmu(env, addr + (2 << DF_WORD), oi, GETPC());
 #endif
+#else
+#if !defined(HOST_WORDS_BIGENDIAN)
+    pwd->w[0] = cpu_ldl_data(env, addr + (0 << DF_WORD));
+    pwd->w[1] = cpu_ldl_data(env, addr + (1 << DF_WORD));
+    pwd->w[2] = cpu_ldl_data(env, addr + (2 << DF_WORD));
+    pwd->w[3] = cpu_ldl_data(env, addr + (3 << DF_WORD));
+#else
+    pwd->w[0] = cpu_ldl_data(env, addr + (1 << DF_WORD));
+    pwd->w[1] = cpu_ldl_data(env, addr + (0 << DF_WORD));
+    pwd->w[2] = cpu_ldl_data(env, addr + (3 << DF_WORD));
+    pwd->w[3] = cpu_ldl_data(env, addr + (2 << DF_WORD));
+#endif
+#endif
+}
+
+void helper_msa_ld_d(CPUMIPSState *env, uint32_t wd,
+                     target_ulong addr)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    MEMOP_IDX(DF_DOUBLE)
+#if !defined(CONFIG_USER_ONLY)
+    pwd->d[0] = helper_ret_ldq_mmu(env, addr + (0 << DF_DOUBLE), oi, GETPC());
+    pwd->d[1] = helper_ret_ldq_mmu(env, addr + (1 << DF_DOUBLE), oi, GETPC());
+#else
+    pwd->d[0] = cpu_ldq_data(env, addr + (0 << DF_DOUBLE));
+    pwd->d[1] = cpu_ldq_data(env, addr + (1 << DF_DOUBLE));
+#endif
+}
 
 #define MSA_PAGESPAN(x) \
         ((((x) & ~TARGET_PAGE_MASK) + MSA_WRLEN/8 - 1) >= TARGET_PAGE_SIZE)
@@ -4402,31 +4550,191 @@ static inline void ensure_writable_pages(CPUMIPSState *env,
 #endif
 }
 
-#define MSA_ST_DF(DF, TYPE, ST_INSN, ...)                               \
-void helper_msa_st_ ## TYPE(CPUMIPSState *env, uint32_t wd,             \
-                            target_ulong addr)                          \
-{                                                                       \
-    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);                          \
-    int mmu_idx = cpu_mmu_index(env, false);				\
-    int i;                                                              \
-    MEMOP_IDX(DF)                                                       \
-    ensure_writable_pages(env, addr, mmu_idx, GETPC());                 \
-    for (i = 0; i < DF_ELEMENTS(DF); i++) {                             \
-        ST_INSN(env, addr + (i << DF), pwd->TYPE[i], ##__VA_ARGS__);    \
-    }                                                                   \
+void helper_msa_st_b(CPUMIPSState *env, uint32_t wd,
+                     target_ulong addr)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    int mmu_idx = cpu_mmu_index(env, false);
+
+    MEMOP_IDX(DF_BYTE)
+    ensure_writable_pages(env, addr, mmu_idx, GETPC());
+#if !defined(CONFIG_USER_ONLY)
+#if !defined(HOST_WORDS_BIGENDIAN)
+    helper_ret_stb_mmu(env, addr + (0  << DF_BYTE), pwd->b[0],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (1  << DF_BYTE), pwd->b[1],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (2  << DF_BYTE), pwd->b[2],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (3  << DF_BYTE), pwd->b[3],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (4  << DF_BYTE), pwd->b[4],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (5  << DF_BYTE), pwd->b[5],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (6  << DF_BYTE), pwd->b[6],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (7  << DF_BYTE), pwd->b[7],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (8  << DF_BYTE), pwd->b[8],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (9  << DF_BYTE), pwd->b[9],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (10 << DF_BYTE), pwd->b[10], oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (11 << DF_BYTE), pwd->b[11], oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (12 << DF_BYTE), pwd->b[12], oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (13 << DF_BYTE), pwd->b[13], oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (14 << DF_BYTE), pwd->b[14], oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (15 << DF_BYTE), pwd->b[15], oi, GETPC());
+#else
+    helper_ret_stb_mmu(env, addr + (7  << DF_BYTE), pwd->b[0],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (6  << DF_BYTE), pwd->b[1],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (5  << DF_BYTE), pwd->b[2],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (4  << DF_BYTE), pwd->b[3],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (3  << DF_BYTE), pwd->b[4],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (2  << DF_BYTE), pwd->b[5],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (1  << DF_BYTE), pwd->b[6],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (0  << DF_BYTE), pwd->b[7],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (15 << DF_BYTE), pwd->b[8],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (14 << DF_BYTE), pwd->b[9],  oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (13 << DF_BYTE), pwd->b[10], oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (12 << DF_BYTE), pwd->b[11], oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (11 << DF_BYTE), pwd->b[12], oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (10 << DF_BYTE), pwd->b[13], oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (9  << DF_BYTE), pwd->b[14], oi, GETPC());
+    helper_ret_stb_mmu(env, addr + (8  << DF_BYTE), pwd->b[15], oi, GETPC());
+#endif
+#else
+#if !defined(HOST_WORDS_BIGENDIAN)
+    cpu_stb_data(env, addr + (0  << DF_BYTE), pwd->b[0]);
+    cpu_stb_data(env, addr + (1  << DF_BYTE), pwd->b[1]);
+    cpu_stb_data(env, addr + (2  << DF_BYTE), pwd->b[2]);
+    cpu_stb_data(env, addr + (3  << DF_BYTE), pwd->b[3]);
+    cpu_stb_data(env, addr + (4  << DF_BYTE), pwd->b[4]);
+    cpu_stb_data(env, addr + (5  << DF_BYTE), pwd->b[5]);
+    cpu_stb_data(env, addr + (6  << DF_BYTE), pwd->b[6]);
+    cpu_stb_data(env, addr + (7  << DF_BYTE), pwd->b[7]);
+    cpu_stb_data(env, addr + (8  << DF_BYTE), pwd->b[8]);
+    cpu_stb_data(env, addr + (9  << DF_BYTE), pwd->b[9]);
+    cpu_stb_data(env, addr + (10 << DF_BYTE), pwd->b[10]);
+    cpu_stb_data(env, addr + (11 << DF_BYTE), pwd->b[11]);
+    cpu_stb_data(env, addr + (12 << DF_BYTE), pwd->b[12]);
+    cpu_stb_data(env, addr + (13 << DF_BYTE), pwd->b[13]);
+    cpu_stb_data(env, addr + (14 << DF_BYTE), pwd->b[14]);
+    cpu_stb_data(env, addr + (15 << DF_BYTE), pwd->b[15]);
+#else
+    cpu_stb_data(env, addr + (7  << DF_BYTE), pwd->b[0]);
+    cpu_stb_data(env, addr + (6  << DF_BYTE), pwd->b[1]);
+    cpu_stb_data(env, addr + (5  << DF_BYTE), pwd->b[2]);
+    cpu_stb_data(env, addr + (4  << DF_BYTE), pwd->b[3]);
+    cpu_stb_data(env, addr + (3  << DF_BYTE), pwd->b[4]);
+    cpu_stb_data(env, addr + (2  << DF_BYTE), pwd->b[5]);
+    cpu_stb_data(env, addr + (1  << DF_BYTE), pwd->b[6]);
+    cpu_stb_data(env, addr + (0  << DF_BYTE), pwd->b[7]);
+    cpu_stb_data(env, addr + (15 << DF_BYTE), pwd->b[8]);
+    cpu_stb_data(env, addr + (14 << DF_BYTE), pwd->b[9]);
+    cpu_stb_data(env, addr + (13 << DF_BYTE), pwd->b[10]);
+    cpu_stb_data(env, addr + (12 << DF_BYTE), pwd->b[11]);
+    cpu_stb_data(env, addr + (11 << DF_BYTE), pwd->b[12]);
+    cpu_stb_data(env, addr + (10 << DF_BYTE), pwd->b[13]);
+    cpu_stb_data(env, addr + (9  << DF_BYTE), pwd->b[14]);
+    cpu_stb_data(env, addr + (8  << DF_BYTE), pwd->b[15]);
+#endif
+#endif
+}
+
+void helper_msa_st_h(CPUMIPSState *env, uint32_t wd,
+                     target_ulong addr)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    int mmu_idx = cpu_mmu_index(env, false);
+
+    MEMOP_IDX(DF_HALF)
+    ensure_writable_pages(env, addr, mmu_idx, GETPC());
+#if !defined(CONFIG_USER_ONLY)
+#if !defined(HOST_WORDS_BIGENDIAN)
+    helper_ret_stw_mmu(env, addr + (0 << DF_HALF), pwd->h[0], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (1 << DF_HALF), pwd->h[1], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (2 << DF_HALF), pwd->h[2], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (3 << DF_HALF), pwd->h[3], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (4 << DF_HALF), pwd->h[4], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (5 << DF_HALF), pwd->h[5], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (6 << DF_HALF), pwd->h[6], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (7 << DF_HALF), pwd->h[7], oi, GETPC());
+#else
+    helper_ret_stw_mmu(env, addr + (3 << DF_HALF), pwd->h[0], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (2 << DF_HALF), pwd->h[1], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (1 << DF_HALF), pwd->h[2], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (0 << DF_HALF), pwd->h[3], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (7 << DF_HALF), pwd->h[4], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (6 << DF_HALF), pwd->h[5], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (5 << DF_HALF), pwd->h[6], oi, GETPC());
+    helper_ret_stw_mmu(env, addr + (4 << DF_HALF), pwd->h[7], oi, GETPC());
+#endif
+#else
+#if !defined(HOST_WORDS_BIGENDIAN)
+    cpu_stw_data(env, addr + (0 << DF_HALF), pwd->h[0]);
+    cpu_stw_data(env, addr + (1 << DF_HALF), pwd->h[1]);
+    cpu_stw_data(env, addr + (2 << DF_HALF), pwd->h[2]);
+    cpu_stw_data(env, addr + (3 << DF_HALF), pwd->h[3]);
+    cpu_stw_data(env, addr + (4 << DF_HALF), pwd->h[4]);
+    cpu_stw_data(env, addr + (5 << DF_HALF), pwd->h[5]);
+    cpu_stw_data(env, addr + (6 << DF_HALF), pwd->h[6]);
+    cpu_stw_data(env, addr + (7 << DF_HALF), pwd->h[7]);
+#else
+    cpu_stw_data(env, addr + (3 << DF_HALF), pwd->h[0]);
+    cpu_stw_data(env, addr + (2 << DF_HALF), pwd->h[1]);
+    cpu_stw_data(env, addr + (1 << DF_HALF), pwd->h[2]);
+    cpu_stw_data(env, addr + (0 << DF_HALF), pwd->h[3]);
+    cpu_stw_data(env, addr + (7 << DF_HALF), pwd->h[4]);
+    cpu_stw_data(env, addr + (6 << DF_HALF), pwd->h[5]);
+    cpu_stw_data(env, addr + (5 << DF_HALF), pwd->h[6]);
+    cpu_stw_data(env, addr + (4 << DF_HALF), pwd->h[7]);
+#endif
+#endif
 }
 
+void helper_msa_st_w(CPUMIPSState *env, uint32_t wd,
+                     target_ulong addr)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    int mmu_idx = cpu_mmu_index(env, false);
+
+    MEMOP_IDX(DF_WORD)
+    ensure_writable_pages(env, addr, mmu_idx, GETPC());
 #if !defined(CONFIG_USER_ONLY)
-MSA_ST_DF(DF_BYTE,   b, helper_ret_stb_mmu, oi, GETPC())
-MSA_ST_DF(DF_HALF,   h, helper_ret_stw_mmu, oi, GETPC())
-MSA_ST_DF(DF_WORD,   w, helper_ret_stl_mmu, oi, GETPC())
-MSA_ST_DF(DF_DOUBLE, d, helper_ret_stq_mmu, oi, GETPC())
+#if !defined(HOST_WORDS_BIGENDIAN)
+    helper_ret_stl_mmu(env, addr + (0 << DF_WORD), oi, GETPC(), pwd->w[0]);
+    helper_ret_stl_mmu(env, addr + (1 << DF_WORD), oi, GETPC(), pwd->w[1]);
+    helper_ret_stl_mmu(env, addr + (2 << DF_WORD), oi, GETPC(), pwd->w[2]);
+    helper_ret_stl_mmu(env, addr + (3 << DF_WORD), oi, GETPC(), pwd->w[3]);
 #else
-MSA_ST_DF(DF_BYTE,   b, cpu_stb_data)
-MSA_ST_DF(DF_HALF,   h, cpu_stw_data)
-MSA_ST_DF(DF_WORD,   w, cpu_stl_data)
-MSA_ST_DF(DF_DOUBLE, d, cpu_stq_data)
+    helper_ret_stl_mmu(env, addr + (1 << DF_WORD), oi, GETPC(), pwd->w[0]);
+    helper_ret_stl_mmu(env, addr + (0 << DF_WORD), oi, GETPC(), pwd->w[1]);
+    helper_ret_stl_mmu(env, addr + (3 << DF_WORD), oi, GETPC(), pwd->w[2]);
+    helper_ret_stl_mmu(env, addr + (2 << DF_WORD), oi, GETPC(), pwd->w[3]);
+#endif
+#else
+#if !defined(HOST_WORDS_BIGENDIAN)
+    cpu_stl_data(env, addr + (0 << DF_WORD), pwd->w[0]);
+    cpu_stl_data(env, addr + (1 << DF_WORD), pwd->w[1]);
+    cpu_stl_data(env, addr + (2 << DF_WORD), pwd->w[2]);
+    cpu_stl_data(env, addr + (3 << DF_WORD), pwd->w[3]);
+#else
+    cpu_stl_data(env, addr + (1 << DF_WORD), pwd->w[0]);
+    cpu_stl_data(env, addr + (0 << DF_WORD), pwd->w[1]);
+    cpu_stl_data(env, addr + (3 << DF_WORD), pwd->w[2]);
+    cpu_stl_data(env, addr + (2 << DF_WORD), pwd->w[3]);
+#endif
 #endif
+}
+
+void helper_msa_st_d(CPUMIPSState *env, uint32_t wd,
+                     target_ulong addr)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    int mmu_idx = cpu_mmu_index(env, false);
+
+    MEMOP_IDX(DF_DOUBLE)
+    ensure_writable_pages(env, addr, mmu_idx, GETPC());
+#if !defined(CONFIG_USER_ONLY)
+    helper_ret_stq_mmu(env, addr + (0 << DF_DOUBLE), pwd->d[0], oi, GETPC());
+    helper_ret_stq_mmu(env, addr + (1 << DF_DOUBLE), pwd->d[1], oi, GETPC());
+#else
+    cpu_stq_data(env, addr + (0 << DF_DOUBLE), pwd->d[0]);
+    cpu_stq_data(env, addr + (1 << DF_DOUBLE), pwd->d[1]);
+#endif
+}
 
 void helper_cache(CPUMIPSState *env, target_ulong addr, uint32_t op)
 {