target/mips: msa: Split helpers for <MAX|MIN>_A.<B|H|W|D>

Achieves clearer code and slightly better performance.

Reviewed-by: Aleksandar Rikalo <aleksandar.rikalo@rt-rk.com>
Signed-off-by: Aleksandar Markovic <amarkovic@wavecomp.com>
Message-Id: <1571826227-10583-5-git-send-email-aleksandar.markovic@rt-rk.com>

1 files changed, 146 insertions, 17 deletions

diff --git a/target/mips/msa_helper.c b/target/mips/msa_helper.c
index a2052baa57..3eb0ab1d6e 100644
--- a/target/mips/msa_helper.c
+++ b/target/mips/msa_helper.c
@@ -1736,7 +1736,152 @@ void helper_msa_div_u_d(CPUMIPSState *env,
  * +---------------+----------------------------------------------------------+
  */
 
-/* TODO: insert Int Max Min group helpers here */
+static inline int64_t msa_max_a_df(uint32_t df, int64_t arg1, int64_t arg2)
+{
+    uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
+    uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
+    return abs_arg1 > abs_arg2 ? arg1 : arg2;
+}
+
+void helper_msa_max_a_b(CPUMIPSState *env,
+                        uint32_t wd, uint32_t ws, uint32_t wt)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    wr_t *pws = &(env->active_fpu.fpr[ws].wr);
+    wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
+
+    pwd->b[0]  = msa_max_a_df(DF_BYTE, pws->b[0],  pwt->b[0]);
+    pwd->b[1]  = msa_max_a_df(DF_BYTE, pws->b[1],  pwt->b[1]);
+    pwd->b[2]  = msa_max_a_df(DF_BYTE, pws->b[2],  pwt->b[2]);
+    pwd->b[3]  = msa_max_a_df(DF_BYTE, pws->b[3],  pwt->b[3]);
+    pwd->b[4]  = msa_max_a_df(DF_BYTE, pws->b[4],  pwt->b[4]);
+    pwd->b[5]  = msa_max_a_df(DF_BYTE, pws->b[5],  pwt->b[5]);
+    pwd->b[6]  = msa_max_a_df(DF_BYTE, pws->b[6],  pwt->b[6]);
+    pwd->b[7]  = msa_max_a_df(DF_BYTE, pws->b[7],  pwt->b[7]);
+    pwd->b[8]  = msa_max_a_df(DF_BYTE, pws->b[8],  pwt->b[8]);
+    pwd->b[9]  = msa_max_a_df(DF_BYTE, pws->b[9],  pwt->b[9]);
+    pwd->b[10] = msa_max_a_df(DF_BYTE, pws->b[10], pwt->b[10]);
+    pwd->b[11] = msa_max_a_df(DF_BYTE, pws->b[11], pwt->b[11]);
+    pwd->b[12] = msa_max_a_df(DF_BYTE, pws->b[12], pwt->b[12]);
+    pwd->b[13] = msa_max_a_df(DF_BYTE, pws->b[13], pwt->b[13]);
+    pwd->b[14] = msa_max_a_df(DF_BYTE, pws->b[14], pwt->b[14]);
+    pwd->b[15] = msa_max_a_df(DF_BYTE, pws->b[15], pwt->b[15]);
+}
+
+void helper_msa_max_a_h(CPUMIPSState *env,
+                        uint32_t wd, uint32_t ws, uint32_t wt)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    wr_t *pws = &(env->active_fpu.fpr[ws].wr);
+    wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
+
+    pwd->h[0]  = msa_max_a_df(DF_HALF, pws->h[0],  pwt->h[0]);
+    pwd->h[1]  = msa_max_a_df(DF_HALF, pws->h[1],  pwt->h[1]);
+    pwd->h[2]  = msa_max_a_df(DF_HALF, pws->h[2],  pwt->h[2]);
+    pwd->h[3]  = msa_max_a_df(DF_HALF, pws->h[3],  pwt->h[3]);
+    pwd->h[4]  = msa_max_a_df(DF_HALF, pws->h[4],  pwt->h[4]);
+    pwd->h[5]  = msa_max_a_df(DF_HALF, pws->h[5],  pwt->h[5]);
+    pwd->h[6]  = msa_max_a_df(DF_HALF, pws->h[6],  pwt->h[6]);
+    pwd->h[7]  = msa_max_a_df(DF_HALF, pws->h[7],  pwt->h[7]);
+}
+
+void helper_msa_max_a_w(CPUMIPSState *env,
+                        uint32_t wd, uint32_t ws, uint32_t wt)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    wr_t *pws = &(env->active_fpu.fpr[ws].wr);
+    wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
+
+    pwd->w[0]  = msa_max_a_df(DF_WORD, pws->w[0],  pwt->w[0]);
+    pwd->w[1]  = msa_max_a_df(DF_WORD, pws->w[1],  pwt->w[1]);
+    pwd->w[2]  = msa_max_a_df(DF_WORD, pws->w[2],  pwt->w[2]);
+    pwd->w[3]  = msa_max_a_df(DF_WORD, pws->w[3],  pwt->w[3]);
+}
+
+void helper_msa_max_a_d(CPUMIPSState *env,
+                        uint32_t wd, uint32_t ws, uint32_t wt)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    wr_t *pws = &(env->active_fpu.fpr[ws].wr);
+    wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
+
+    pwd->d[0]  = msa_max_a_df(DF_DOUBLE, pws->d[0],  pwt->d[0]);
+    pwd->d[1]  = msa_max_a_df(DF_DOUBLE, pws->d[1],  pwt->d[1]);
+}
+
+
+static inline int64_t msa_min_a_df(uint32_t df, int64_t arg1, int64_t arg2)
+{
+    uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
+    uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
+    return abs_arg1 < abs_arg2 ? arg1 : arg2;
+}
+
+void helper_msa_min_a_b(CPUMIPSState *env,
+                        uint32_t wd, uint32_t ws, uint32_t wt)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    wr_t *pws = &(env->active_fpu.fpr[ws].wr);
+    wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
+
+    pwd->b[0]  = msa_min_a_df(DF_BYTE, pws->b[0],  pwt->b[0]);
+    pwd->b[1]  = msa_min_a_df(DF_BYTE, pws->b[1],  pwt->b[1]);
+    pwd->b[2]  = msa_min_a_df(DF_BYTE, pws->b[2],  pwt->b[2]);
+    pwd->b[3]  = msa_min_a_df(DF_BYTE, pws->b[3],  pwt->b[3]);
+    pwd->b[4]  = msa_min_a_df(DF_BYTE, pws->b[4],  pwt->b[4]);
+    pwd->b[5]  = msa_min_a_df(DF_BYTE, pws->b[5],  pwt->b[5]);
+    pwd->b[6]  = msa_min_a_df(DF_BYTE, pws->b[6],  pwt->b[6]);
+    pwd->b[7]  = msa_min_a_df(DF_BYTE, pws->b[7],  pwt->b[7]);
+    pwd->b[8]  = msa_min_a_df(DF_BYTE, pws->b[8],  pwt->b[8]);
+    pwd->b[9]  = msa_min_a_df(DF_BYTE, pws->b[9],  pwt->b[9]);
+    pwd->b[10] = msa_min_a_df(DF_BYTE, pws->b[10], pwt->b[10]);
+    pwd->b[11] = msa_min_a_df(DF_BYTE, pws->b[11], pwt->b[11]);
+    pwd->b[12] = msa_min_a_df(DF_BYTE, pws->b[12], pwt->b[12]);
+    pwd->b[13] = msa_min_a_df(DF_BYTE, pws->b[13], pwt->b[13]);
+    pwd->b[14] = msa_min_a_df(DF_BYTE, pws->b[14], pwt->b[14]);
+    pwd->b[15] = msa_min_a_df(DF_BYTE, pws->b[15], pwt->b[15]);
+}
+
+void helper_msa_min_a_h(CPUMIPSState *env,
+                        uint32_t wd, uint32_t ws, uint32_t wt)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    wr_t *pws = &(env->active_fpu.fpr[ws].wr);
+    wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
+
+    pwd->h[0]  = msa_min_a_df(DF_HALF, pws->h[0],  pwt->h[0]);
+    pwd->h[1]  = msa_min_a_df(DF_HALF, pws->h[1],  pwt->h[1]);
+    pwd->h[2]  = msa_min_a_df(DF_HALF, pws->h[2],  pwt->h[2]);
+    pwd->h[3]  = msa_min_a_df(DF_HALF, pws->h[3],  pwt->h[3]);
+    pwd->h[4]  = msa_min_a_df(DF_HALF, pws->h[4],  pwt->h[4]);
+    pwd->h[5]  = msa_min_a_df(DF_HALF, pws->h[5],  pwt->h[5]);
+    pwd->h[6]  = msa_min_a_df(DF_HALF, pws->h[6],  pwt->h[6]);
+    pwd->h[7]  = msa_min_a_df(DF_HALF, pws->h[7],  pwt->h[7]);
+}
+
+void helper_msa_min_a_w(CPUMIPSState *env,
+                        uint32_t wd, uint32_t ws, uint32_t wt)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    wr_t *pws = &(env->active_fpu.fpr[ws].wr);
+    wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
+
+    pwd->w[0]  = msa_min_a_df(DF_WORD, pws->w[0],  pwt->w[0]);
+    pwd->w[1]  = msa_min_a_df(DF_WORD, pws->w[1],  pwt->w[1]);
+    pwd->w[2]  = msa_min_a_df(DF_WORD, pws->w[2],  pwt->w[2]);
+    pwd->w[3]  = msa_min_a_df(DF_WORD, pws->w[3],  pwt->w[3]);
+}
+
+void helper_msa_min_a_d(CPUMIPSState *env,
+                        uint32_t wd, uint32_t ws, uint32_t wt)
+{
+    wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
+    wr_t *pws = &(env->active_fpu.fpr[ws].wr);
+    wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
+
+    pwd->d[0]  = msa_min_a_df(DF_DOUBLE, pws->d[0],  pwt->d[0]);
+    pwd->d[1]  = msa_min_a_df(DF_DOUBLE, pws->d[1],  pwt->d[1]);
+}
 
 
 /*
@@ -2456,20 +2601,6 @@ MSA_TEROP_IMMU_DF(binsli, binsl)
 MSA_TEROP_IMMU_DF(binsri, binsr)
 #undef MSA_TEROP_IMMU_DF
 
-static inline int64_t msa_max_a_df(uint32_t df, int64_t arg1, int64_t arg2)
-{
-    uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
-    uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
-    return abs_arg1 > abs_arg2 ? arg1 : arg2;
-}
-
-static inline int64_t msa_min_a_df(uint32_t df, int64_t arg1, int64_t arg2)
-{
-    uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
-    uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
-    return abs_arg1 < abs_arg2 ? arg1 : arg2;
-}
-
 static inline int64_t msa_add_a_df(uint32_t df, int64_t arg1, int64_t arg2)
 {
     uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
@@ -2773,8 +2904,6 @@ MSA_BINOP_DF(max_s)
 MSA_BINOP_DF(max_u)
 MSA_BINOP_DF(min_s)
 MSA_BINOP_DF(min_u)
-MSA_BINOP_DF(max_a)
-MSA_BINOP_DF(min_a)
 MSA_BINOP_DF(add_a)
 MSA_BINOP_DF(adds_a)
 MSA_BINOP_DF(adds_s)