[PATCH v8 06/14] target/mips: msa: Split helpers for DPSUB_U.<H|W|D>

[Aleksandar Markovic <aleksandar.qemu.devel@gmail.com>]

Achieves clearer code and slightly better performance.

Signed-off-by: Aleksandar Markovic <aleksandar.qemu.devel@gmail.com>
---
 target/mips/helper.h     |  4 ++-
 target/mips/msa_helper.c | 67 ++++++++++++++++++++++++++++++++--------
 target/mips/translate.c  | 12 ++++++-
 3 files changed, 68 insertions(+), 15 deletions(-)

diff --git a/target/mips/helper.h b/target/mips/helper.h
index 2de14542cd..575f4a524c 100644
--- a/target/mips/helper.h
+++ b/target/mips/helper.h
@@ -1090,7 +1090,9 @@ DEF_HELPER_4(msa_dpadd_u_d, void, env, i32, i32, i32)
 DEF_HELPER_4(msa_dpsub_s_h, void, env, i32, i32, i32)
 DEF_HELPER_4(msa_dpsub_s_w, void, env, i32, i32, i32)
 DEF_HELPER_4(msa_dpsub_s_d, void, env, i32, i32, i32)
-DEF_HELPER_5(msa_dpsub_u_df, void, env, i32, i32, i32, i32)
+DEF_HELPER_4(msa_dpsub_u_h, void, env, i32, i32, i32)
+DEF_HELPER_4(msa_dpsub_u_w, void, env, i32, i32, i32)
+DEF_HELPER_4(msa_dpsub_u_d, void, env, i32, i32, i32)
 DEF_HELPER_5(msa_sld_df, void, env, i32, i32, i32, i32)
 DEF_HELPER_5(msa_splat_df, void, env, i32, i32, i32, i32)
 DEF_HELPER_5(msa_vshf_df, void, env, i32, i32, i32, i32)
diff --git a/target/mips/msa_helper.c b/target/mips/msa_helper.c
index 934f705c1e..33d5251a6b 100644
--- a/target/mips/msa_helper.c
+++ b/target/mips/msa_helper.c
@@ -2398,6 +2398,60 @@ void helper_msa_dpsub_s_d(CPUMIPSState *env,
 }
 
 
+static inline int64_t msa_dpsub_u_df(uint32_t df, int64_t dest, int64_t arg1,
+                                     int64_t arg2)
+{
+    int64_t even_arg1;
+    int64_t even_arg2;
+    int64_t odd_arg1;
+    int64_t odd_arg2;
+    UNSIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
+    UNSIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
+    return dest - ((even_arg1 * even_arg2) + (odd_arg1 * odd_arg2));
+}
+
+void helper_msa_dpsub_u_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_dpsub_u_df(DF_HALF, pwd->h[0],  pws->h[0],  pwt->h[0]);
+    pwd->h[1]  = msa_dpsub_u_df(DF_HALF, pwd->h[1],  pws->h[1],  pwt->h[1]);
+    pwd->h[2]  = msa_dpsub_u_df(DF_HALF, pwd->h[2],  pws->h[2],  pwt->h[2]);
+    pwd->h[3]  = msa_dpsub_u_df(DF_HALF, pwd->h[3],  pws->h[3],  pwt->h[3]);
+    pwd->h[4]  = msa_dpsub_u_df(DF_HALF, pwd->h[4],  pws->h[4],  pwt->h[4]);
+    pwd->h[5]  = msa_dpsub_u_df(DF_HALF, pwd->h[5],  pws->h[5],  pwt->h[5]);
+    pwd->h[6]  = msa_dpsub_u_df(DF_HALF, pwd->h[6],  pws->h[6],  pwt->h[6]);
+    pwd->h[7]  = msa_dpsub_u_df(DF_HALF, pwd->h[7],  pws->h[7],  pwt->h[7]);
+}
+
+void helper_msa_dpsub_u_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_dpsub_u_df(DF_WORD, pwd->w[0],  pws->w[0],  pwt->w[0]);
+    pwd->w[1]  = msa_dpsub_u_df(DF_WORD, pwd->w[1],  pws->w[1],  pwt->w[1]);
+    pwd->w[2]  = msa_dpsub_u_df(DF_WORD, pwd->w[2],  pws->w[2],  pwt->w[2]);
+    pwd->w[3]  = msa_dpsub_u_df(DF_WORD, pwd->w[3],  pws->w[3],  pwt->w[3]);
+}
+
+void helper_msa_dpsub_u_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_dpsub_u_df(DF_DOUBLE, pwd->d[0],  pws->d[0],  pwt->d[0]);
+    pwd->d[1]  = msa_dpsub_u_df(DF_DOUBLE, pwd->d[1],  pws->d[1],  pwt->d[1]);
+}
+
+
 /*
  * Int Max Min
  * -----------
@@ -5117,18 +5171,6 @@ void helper_msa_sld_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
     msa_sld_df(df, pwd, pws, env->active_tc.gpr[rt]);
 }
 
-static inline int64_t msa_dpsub_u_df(uint32_t df, int64_t dest, int64_t arg1,
-                                     int64_t arg2)
-{
-    int64_t even_arg1;
-    int64_t even_arg2;
-    int64_t odd_arg1;
-    int64_t odd_arg2;
-    UNSIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
-    UNSIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
-    return dest - ((even_arg1 * even_arg2) + (odd_arg1 * odd_arg2));
-}
-
 static inline int64_t msa_madd_q_df(uint32_t df, int64_t dest, int64_t arg1,
                                     int64_t arg2)
 {
@@ -5255,7 +5297,6 @@ void helper_msa_ ## func ## _df(CPUMIPSState *env, uint32_t df, uint32_t wd,  \
     }                                                                         \
 }
 
-MSA_TEROP_DF(dpsub_u)
 MSA_TEROP_DF(binsl)
 MSA_TEROP_DF(binsr)
 MSA_TEROP_DF(madd_q)
diff --git a/target/mips/translate.c b/target/mips/translate.c
index 2576905e5b..3dda242643 100644
--- a/target/mips/translate.c
+++ b/target/mips/translate.c
@@ -29438,7 +29438,17 @@ static void gen_msa_3r(CPUMIPSState *env, DisasContext *ctx)
             }
             break;
         case OPC_DPSUB_U_df:
-            gen_helper_msa_dpsub_u_df(cpu_env, tdf, twd, tws, twt);
+            switch (df) {
+            case DF_HALF:
+                gen_helper_msa_dpsub_u_h(cpu_env, twd, tws, twt);
+                break;
+            case DF_WORD:
+                gen_helper_msa_dpsub_u_w(cpu_env, twd, tws, twt);
+                break;
+            case DF_DOUBLE:
+                gen_helper_msa_dpsub_u_d(cpu_env, twd, tws, twt);
+                break;
+            }
             break;
         }
         break;
-- 
2.20.1