[PULL 05/18] target/mips: msa: Split helpers for DPADD_S.<H|W|D>

[Aleksandar Markovic]

Achieves clearer code and slightly better performance.

Reviewed-by: Aleksandar Rikalo <aleksandar.rikalo@syrmia.com>
Signed-off-by: Aleksandar Markovic <aleksandar.qemu.devel@gmail.com>
Message-Id: <20200613152133.8964-4-aleksandar.qemu.devel@gmail.com>
---
 target/mips/helper.h     |  4 ++-
 target/mips/msa_helper.c | 90 ++++++++++++++++++++++++++++++++++--------------
 target/mips/translate.c  | 12 ++++++-
 3 files changed, 78 insertions(+), 28 deletions(-)

diff --git a/target/mips/helper.h b/target/mips/helper.h
index 7ca0036..16f2d53 100644
--- a/target/mips/helper.h
+++ b/target/mips/helper.h
@@ -1081,7 +1081,9 @@ DEF_HELPER_5(msa_subsuu_s_df, void, env, i32, i32, i32, 
i32)
 DEF_HELPER_5(msa_mulv_df, void, env, i32, i32, i32, i32)
 DEF_HELPER_5(msa_dotp_s_df, void, env, i32, i32, i32, i32)
 DEF_HELPER_5(msa_dotp_u_df, void, env, i32, i32, i32, i32)
-DEF_HELPER_5(msa_dpadd_s_df, void, env, i32, i32, i32, i32)
+DEF_HELPER_4(msa_dpadd_s_h, void, env, i32, i32, i32)
+DEF_HELPER_4(msa_dpadd_s_w, void, env, i32, i32, i32)
+DEF_HELPER_4(msa_dpadd_s_d, void, env, i32, i32, i32)
 DEF_HELPER_5(msa_dpadd_u_df, void, env, i32, i32, i32, i32)
 DEF_HELPER_5(msa_dpsub_s_df, void, env, i32, i32, i32, i32)
 DEF_HELPER_5(msa_dpsub_u_df, void, env, i32, i32, i32, i32)
diff --git a/target/mips/msa_helper.c b/target/mips/msa_helper.c
index 2b54de0..086b56f 100644
--- a/target/mips/msa_helper.c
+++ b/target/mips/msa_helper.c
@@ -2224,7 +2224,70 @@ void helper_msa_div_u_d(CPUMIPSState *env,
  * +---------------+----------------------------------------------------------+
  */
 
-/* TODO: insert Int Dot Product group helpers here */
+#define SIGNED_EXTRACT(e, o, a, df)     \
+    do {                                \
+        e = SIGNED_EVEN(a, df);         \
+        o = SIGNED_ODD(a, df);          \
+    } while (0)
+
+#define UNSIGNED_EXTRACT(e, o, a, df)   \
+    do {                                \
+        e = UNSIGNED_EVEN(a, df);       \
+        o = UNSIGNED_ODD(a, df);        \
+    } while (0)
+
+static inline int64_t msa_dpadd_s_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;
+    SIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
+    SIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
+    return dest + (even_arg1 * even_arg2) + (odd_arg1 * odd_arg2);
+}
+
+void helper_msa_dpadd_s_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_dpadd_s_df(DF_HALF, pwd->h[0],  pws->h[0],  pwt->h[0]);
+    pwd->h[1]  = msa_dpadd_s_df(DF_HALF, pwd->h[1],  pws->h[1],  pwt->h[1]);
+    pwd->h[2]  = msa_dpadd_s_df(DF_HALF, pwd->h[2],  pws->h[2],  pwt->h[2]);
+    pwd->h[3]  = msa_dpadd_s_df(DF_HALF, pwd->h[3],  pws->h[3],  pwt->h[3]);
+    pwd->h[4]  = msa_dpadd_s_df(DF_HALF, pwd->h[4],  pws->h[4],  pwt->h[4]);
+    pwd->h[5]  = msa_dpadd_s_df(DF_HALF, pwd->h[5],  pws->h[5],  pwt->h[5]);
+    pwd->h[6]  = msa_dpadd_s_df(DF_HALF, pwd->h[6],  pws->h[6],  pwt->h[6]);
+    pwd->h[7]  = msa_dpadd_s_df(DF_HALF, pwd->h[7],  pws->h[7],  pwt->h[7]);
+}
+
+void helper_msa_dpadd_s_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_dpadd_s_df(DF_WORD, pwd->w[0],  pws->w[0],  pwt->w[0]);
+    pwd->w[1]  = msa_dpadd_s_df(DF_WORD, pwd->w[1],  pws->w[1],  pwt->w[1]);
+    pwd->w[2]  = msa_dpadd_s_df(DF_WORD, pwd->w[2],  pws->w[2],  pwt->w[2]);
+    pwd->w[3]  = msa_dpadd_s_df(DF_WORD, pwd->w[3],  pws->w[3],  pwt->w[3]);
+}
+
+void helper_msa_dpadd_s_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_dpadd_s_df(DF_DOUBLE, pwd->d[0],  pws->d[0],  pwt->d[0]);
+    pwd->d[1]  = msa_dpadd_s_df(DF_DOUBLE, pwd->d[1],  pws->d[1],  pwt->d[1]);
+}
 
 
 /*
@@ -4785,18 +4848,6 @@ static inline int64_t msa_mulv_df(uint32_t df, int64_t 
arg1, int64_t arg2)
     return arg1 * arg2;
 }
 
-#define SIGNED_EXTRACT(e, o, a, df)     \
-    do {                                \
-        e = SIGNED_EVEN(a, df);         \
-        o = SIGNED_ODD(a, df);          \
-    } while (0)
-
-#define UNSIGNED_EXTRACT(e, o, a, df)   \
-    do {                                \
-        e = UNSIGNED_EVEN(a, df);       \
-        o = UNSIGNED_ODD(a, df);        \
-    } while (0)
-
 static inline int64_t msa_dotp_s_df(uint32_t df, int64_t arg1, int64_t arg2)
 {
     int64_t even_arg1;
@@ -4958,18 +5009,6 @@ 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_dpadd_s_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;
-    SIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
-    SIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
-    return dest + (even_arg1 * even_arg2) + (odd_arg1 * odd_arg2);
-}
-
 static inline int64_t msa_dpadd_u_df(uint32_t df, int64_t dest, int64_t arg1,
                                      int64_t arg2)
 {
@@ -5132,7 +5171,6 @@ void helper_msa_ ## func ## _df(CPUMIPSState *env, 
uint32_t df, uint32_t wd,  \
     }                                                                         \
 }
 
-MSA_TEROP_DF(dpadd_s)
 MSA_TEROP_DF(dpadd_u)
 MSA_TEROP_DF(dpsub_s)
 MSA_TEROP_DF(dpsub_u)
diff --git a/target/mips/translate.c b/target/mips/translate.c
index 7388a16..47ac4ea 100644
--- a/target/mips/translate.c
+++ b/target/mips/translate.c
@@ -29400,7 +29400,17 @@ static void gen_msa_3r(CPUMIPSState *env, DisasContext 
*ctx)
             gen_helper_msa_dotp_u_df(cpu_env, tdf, twd, tws, twt);
             break;
         case OPC_DPADD_S_df:
-            gen_helper_msa_dpadd_s_df(cpu_env, tdf, twd, tws, twt);
+            switch (df) {
+            case DF_HALF:
+                gen_helper_msa_dpadd_s_h(cpu_env, twd, tws, twt);
+                break;
+            case DF_WORD:
+                gen_helper_msa_dpadd_s_w(cpu_env, twd, tws, twt);
+                break;
+            case DF_DOUBLE:
+                gen_helper_msa_dpadd_s_d(cpu_env, twd, tws, twt);
+                break;
+            }
             break;
         case OPC_DPADD_U_df:
             gen_helper_msa_dpadd_u_df(cpu_env, tdf, twd, tws, twt);
-- 
2.7.4