[Qemu-devel] [PATCH v2 0/4] target/mips: Optimize support for certain MSA instructions

[Qemu-devel] [PATCH v2 0/4] target/mips: Optimize support for certain MSA instructions

[Mateja Marjanovic]

From: Mateja Marjanovic <Mateja.Marjanovic@rt-rk.com>

This series optimizes the support for certain MSA instructions.

v2:
  -Fixed indentation in two places
  -Fixed bugs for the cases when the destination register and
  one of the source registers are the same

Mateja Marjanovic (4):
  target/mips: Optimize support for MSA instructions ILVEV.<B|H|W|D>
  target/mips: Optimize support for MSA instructions ILVOD.<B|H|W|D>
  target/mips: Optimize support for MSA instructions ILVL.<B|H|W|D>
  target/mips: Optimize support for MSA instructions ILVR.<B|H|W|D>

 target/mips/msa_helper.c | 236 ++++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 205 insertions(+), 31 deletions(-)

-- 
2.7.4

[Qemu-devel] [PATCH v2 1/4] target/mips: Optimize support for MSA instructions ILVEV.<B|H|W|D>

[Mateja Marjanovic]

From: Mateja Marjanovic <Mateja.Marjanovic@rt-rk.com>

Optimize support for MSA instructions ILVEV.B, ILVEV.H, ILVEV.W, and
ILVEV.D.

Optimization is done by eliminating loops, and explicitly assigning
desired values to individual data elements. Performance measurement
is done by executing the instructions large number of times on a
computer with Intel Core i7-3770 CPU @ 3.40GHz×8.

Measured time before optimization:
  ILVEV.B:  119.02 ms
  ILVEV.H:   94.16 ms
  ILVEV.W:  120.97 ms
  ILVEV.D:   42.99 ms

Measured time after optimization:
  ILVEV.B:   61.81 ms
  ILVEV.H:   42.78 ms
  ILVEV.W:   39.47 ms
  ILVEV.D:   39.11 ms

Signed-off-by: Mateja Marjanovic <mateja.marjanovic@rt-rk.com>
---
 target/mips/msa_helper.c | 60 +++++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 52 insertions(+), 8 deletions(-)

diff --git a/target/mips/msa_helper.c b/target/mips/msa_helper.c
index c74e3cd..49332e6 100644
--- a/target/mips/msa_helper.c
+++ b/target/mips/msa_helper.c
@@ -1198,14 +1198,6 @@ MSA_FN_DF(ilvl_df)
 MSA_FN_DF(ilvr_df)
 #undef MSA_DO
 
-#define MSA_DO(DF)                      \
-    do {                                \
-        pwx->DF[2*i]   = pwt->DF[2*i];  \
-        pwx->DF[2*i+1] = pws->DF[2*i];  \
-    } while (0)
-MSA_FN_DF(ilvev_df)
-#undef MSA_DO
-
 #define MSA_DO(DF)                          \
     do {                                    \
         pwx->DF[2*i]   = pwt->DF[2*i+1];    \
@@ -1230,6 +1222,58 @@ MSA_FN_DF(vshf_df)
 #undef MSA_LOOP_COND
 #undef MSA_FN_DF
 
+
+void helper_msa_ilvev_df(CPUMIPSState *env, uint32_t df, 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);
+
+    switch (df) {
+    case DF_BYTE:
+        pwd->b[15] = pws->b[14];
+        pwd->b[14] = pwt->b[14];
+        pwd->b[13] = pws->b[12];
+        pwd->b[12] = pwt->b[12];
+        pwd->b[11] = pws->b[10];
+        pwd->b[10] = pwt->b[10];
+        pwd->b[9]  = pws->b[8];
+        pwd->b[8]  = pwt->b[8];
+        pwd->b[7]  = pws->b[6];
+        pwd->b[6]  = pwt->b[6];
+        pwd->b[5]  = pws->b[4];
+        pwd->b[4]  = pwt->b[4];
+        pwd->b[3]  = pws->b[2];
+        pwd->b[2]  = pwt->b[2];
+        pwd->b[1]  = pws->b[0];
+        pwd->b[0]  = pwt->b[0];
+        break;
+    case DF_HALF:
+        pwd->h[7] = pws->h[6];
+        pwd->h[6] = pwt->h[6];
+        pwd->h[5] = pws->h[4];
+        pwd->h[4] = pwt->h[4];
+        pwd->h[3] = pws->h[2];
+        pwd->h[2] = pwt->h[2];
+        pwd->h[1] = pws->h[0];
+        pwd->h[0] = pwt->h[0];
+        break;
+    case DF_WORD:
+        pwd->w[3] = pws->w[2];
+        pwd->w[2] = pwt->w[2];
+        pwd->w[1] = pws->w[0];
+        pwd->w[0] = pwt->w[0];
+        break;
+    case DF_DOUBLE:
+        pwd->d[1] = pws->d[0];
+        pwd->d[0] = pwt->d[0];
+        break;
+    default:
+        assert(0);
+    }
+}
+
 void helper_msa_sldi_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
                         uint32_t ws, uint32_t n)
 {
-- 
2.7.4

[Qemu-devel] [PATCH v2 2/4] target/mips: Optimize support for MSA instructions ILVOD.<B|H|W|D>

[Mateja Marjanovic]

From: Mateja Marjanovic <Mateja.Marjanovic@rt-rk.com>

Optimize support for MSA instructions ILVOD.B, ILVOD.H, ILVOD.W, and
ILVOD.D.

Optimization is done by eliminating loops, and explicitly assigning
desired values to individual data elements. Performance measurement
is done by executing the instructions large number of times on a
computer with Intel Core i7-3770 CPU @ 3.40GHz×8.

Measured time before optimization:
  ILVOD.B:  118.42 ms
  ILVOD.H:   93.45 ms
  ILVOD.W:  119.89 ms
  ILVOD.D:   46.00 ms

Measured time after optimization:
  ILVOD.B:   60.08 ms
  ILVOD.H:   41.52 ms
  ILVOD.W:   36.44 ms
  ILVOD.D:   36.40 ms

Signed-off-by: Mateja Marjanovic <mateja.marjanovic@rt-rk.com>
---
 target/mips/msa_helper.c | 58 ++++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 51 insertions(+), 7 deletions(-)

diff --git a/target/mips/msa_helper.c b/target/mips/msa_helper.c
index 49332e6..2426338 100644
--- a/target/mips/msa_helper.c
+++ b/target/mips/msa_helper.c
@@ -1198,13 +1198,6 @@ MSA_FN_DF(ilvl_df)
 MSA_FN_DF(ilvr_df)
 #undef MSA_DO
 
-#define MSA_DO(DF)                          \
-    do {                                    \
-        pwx->DF[2*i]   = pwt->DF[2*i+1];    \
-        pwx->DF[2*i+1] = pws->DF[2*i+1];    \
-    } while (0)
-MSA_FN_DF(ilvod_df)
-#undef MSA_DO
 #undef MSA_LOOP_COND
 
 #define MSA_LOOP_COND(DF) \
@@ -1274,6 +1267,57 @@ void helper_msa_ilvev_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
     }
 }
 
+void helper_msa_ilvod_df(CPUMIPSState *env, uint32_t df, 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);
+
+    switch (df) {
+    case DF_BYTE:
+        pwd->b[0]  = pwt->b[1];
+        pwd->b[1]  = pws->b[1];
+        pwd->b[2]  = pwt->b[3];
+        pwd->b[3]  = pws->b[3];
+        pwd->b[4]  = pwt->b[5];
+        pwd->b[5]  = pws->b[5];
+        pwd->b[6]  = pwt->b[7];
+        pwd->b[7]  = pws->b[7];
+        pwd->b[8]  = pwt->b[9];
+        pwd->b[9]  = pws->b[9];
+        pwd->b[10] = pwt->b[11];
+        pwd->b[11] = pws->b[11];
+        pwd->b[12] = pwt->b[13];
+        pwd->b[13] = pws->b[13];
+        pwd->b[14] = pwt->b[15];
+        pwd->b[15] = pws->b[15];
+        break;
+    case DF_HALF:
+        pwd->h[0] = pwt->h[1];
+        pwd->h[1] = pws->h[1];
+        pwd->h[2] = pwt->h[3];
+        pwd->h[3] = pws->h[3];
+        pwd->h[4] = pwt->h[5];
+        pwd->h[5] = pws->h[5];
+        pwd->h[6] = pwt->h[7];
+        pwd->h[7] = pws->h[7];
+        break;
+    case DF_WORD:
+        pwd->w[0] = pwt->w[1];
+        pwd->w[1] = pws->w[1];
+        pwd->w[2] = pwt->w[3];
+        pwd->w[3] = pws->w[3];
+        break;
+    case DF_DOUBLE:
+        pwd->d[0] = pwt->d[1];
+        pwd->d[1] = pws->d[1];
+        break;
+    default:
+        assert(0);
+    }
+}
+
 void helper_msa_sldi_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
                         uint32_t ws, uint32_t n)
 {
-- 
2.7.4

[Qemu-devel] [PATCH v2 3/4] target/mips: Optimize support for MSA instructions ILVL.<B|H|W|D>

[Mateja Marjanovic]

From: Mateja Marjanovic <Mateja.Marjanovic@rt-rk.com>

Optimize support for MSA instructions ILVL.B, ILVL.H, ILVL.W, and
ILVL.D.

Optimization is done by eliminating loops, and explicitly assigning
desired values to individual data elements. Performance measurement
is done by executing the instructions large number of times on a
computer with Intel Core i7-3770 CPU @ 3.40GHz×8.

Measured time before optimization:
  ILVL.B:  114.31 ms
  ILVL.H:   92.74 ms
  ILVL.W:  123.96 ms
  ILVL.D:   39.45 ms

Measured time after optimization:
  ILVL.B:   60.18 ms
  ILVL.H:   40.38 ms
  ILVL.W:   38.98 ms
  ILVL.D:   36.48 ms

Signed-off-by: Mateja Marjanovic <mateja.marjanovic@rt-rk.com>
---
 target/mips/msa_helper.c | 59 +++++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 51 insertions(+), 8 deletions(-)

diff --git a/target/mips/msa_helper.c b/target/mips/msa_helper.c
index 2426338..a91f5a6 100644
--- a/target/mips/msa_helper.c
+++ b/target/mips/msa_helper.c
@@ -1184,14 +1184,6 @@ MSA_FN_DF(pckod_df)
 
 #define MSA_DO(DF)                      \
     do {                                \
-        pwx->DF[2*i]   = L##DF(pwt, i); \
-        pwx->DF[2*i+1] = L##DF(pws, i); \
-    } while (0)
-MSA_FN_DF(ilvl_df)
-#undef MSA_DO
-
-#define MSA_DO(DF)                      \
-    do {                                \
         pwx->DF[2*i]   = R##DF(pwt, i); \
         pwx->DF[2*i+1] = R##DF(pws, i); \
     } while (0)
@@ -1318,6 +1310,57 @@ void helper_msa_ilvod_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
     }
 }
 
+void helper_msa_ilvl_df(CPUMIPSState *env, uint32_t df, 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);
+
+    switch (df) {
+    case DF_BYTE:
+        pwd->b[0]  = pwt->b[8];
+        pwd->b[1]  = pws->b[8];
+        pwd->b[2]  = pwt->b[9];
+        pwd->b[3]  = pws->b[9];
+        pwd->b[4]  = pwt->b[10];
+        pwd->b[5]  = pws->b[10];
+        pwd->b[6]  = pwt->b[11];
+        pwd->b[7]  = pws->b[11];
+        pwd->b[8]  = pwt->b[12];
+        pwd->b[9]  = pws->b[12];
+        pwd->b[10] = pwt->b[13];
+        pwd->b[11] = pws->b[13];
+        pwd->b[12] = pwt->b[14];
+        pwd->b[13] = pws->b[14];
+        pwd->b[14] = pwt->b[15];
+        pwd->b[15] = pws->b[15];
+        break;
+    case DF_HALF:
+        pwd->h[0] = pwt->h[4];
+        pwd->h[1] = pws->h[4];
+        pwd->h[2] = pwt->h[5];
+        pwd->h[3] = pws->h[5];
+        pwd->h[4] = pwt->h[6];
+        pwd->h[5] = pws->h[6];
+        pwd->h[6] = pwt->h[7];
+        pwd->h[7] = pws->h[7];
+        break;
+    case DF_WORD:
+        pwd->w[0] = pwt->w[2];
+        pwd->w[1] = pws->w[2];
+        pwd->w[2] = pwt->w[3];
+        pwd->w[3] = pws->w[3];
+        break;
+    case DF_DOUBLE:
+        pwd->d[0] = pwt->d[1];
+        pwd->d[1] = pws->d[1];
+        break;
+    default:
+        assert(0);
+    }
+}
+
 void helper_msa_sldi_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
                         uint32_t ws, uint32_t n)
 {
-- 
2.7.4

[Qemu-devel] [PATCH v2 4/4] target/mips: Optimize support for MSA instructions ILVR.<B|H|W|D>

[Mateja Marjanovic]

From: Mateja Marjanovic <Mateja.Marjanovic@rt-rk.com>

Optimize support for MSA instructions ILVR.B, ILVR.H, ILVR.W, and
ILVR.D.

Optimization is done by eliminating loops, and explicitly assigning
desired values to individual data elements. Performance measurement
is done by executing the instructions large number of times on a
computer with Intel Core i7-3770 CPU @ 3.40GHz×8.

Measured time before optimization:
  ILVR.B:  115.84 ms
  ILVR.H:   94.20 ms
  ILVR.W:  121.12 ms
  ILVR.D:   41.36 ms

Measured time after optimization:
  ILVR.B:   61.06 ms
  ILVR.H:   43.03 ms
  ILVR.W:   39.21 ms
  ILVR.D:   39.18 ms

Signed-off-by: Mateja Marjanovic <mateja.marjanovic@rt-rk.com>
---
 target/mips/msa_helper.c | 59 +++++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 51 insertions(+), 8 deletions(-)

diff --git a/target/mips/msa_helper.c b/target/mips/msa_helper.c
index a91f5a6..4c7ec05 100644
--- a/target/mips/msa_helper.c
+++ b/target/mips/msa_helper.c
@@ -1182,14 +1182,6 @@ MSA_FN_DF(pckev_df)
 MSA_FN_DF(pckod_df)
 #undef MSA_DO
 
-#define MSA_DO(DF)                      \
-    do {                                \
-        pwx->DF[2*i]   = R##DF(pwt, i); \
-        pwx->DF[2*i+1] = R##DF(pws, i); \
-    } while (0)
-MSA_FN_DF(ilvr_df)
-#undef MSA_DO
-
 #undef MSA_LOOP_COND
 
 #define MSA_LOOP_COND(DF) \
@@ -1361,6 +1353,57 @@ void helper_msa_ilvl_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
     }
 }
 
+void helper_msa_ilvr_df(CPUMIPSState *env, uint32_t df, 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);
+
+    switch (df) {
+    case DF_BYTE:
+        pwd->b[15] = pws->b[7];
+        pwd->b[14] = pwt->b[7];
+        pwd->b[13] = pws->b[6];
+        pwd->b[12] = pwt->b[6];
+        pwd->b[11] = pws->b[5];
+        pwd->b[10] = pwt->b[5];
+        pwd->b[9]  = pws->b[4];
+        pwd->b[8]  = pwt->b[4];
+        pwd->b[7]  = pws->b[3];
+        pwd->b[6]  = pwt->b[3];
+        pwd->b[5]  = pws->b[2];
+        pwd->b[4]  = pwt->b[2];
+        pwd->b[3]  = pws->b[1];
+        pwd->b[2]  = pwt->b[1];
+        pwd->b[1]  = pws->b[0];
+        pwd->b[0]  = pwt->b[0];
+        break;
+    case DF_HALF:
+        pwd->h[7] = pws->h[3];
+        pwd->h[6] = pwt->h[3];
+        pwd->h[5] = pws->h[2];
+        pwd->h[4] = pwt->h[2];
+        pwd->h[3] = pws->h[1];
+        pwd->h[2] = pwt->h[1];
+        pwd->h[1] = pws->h[0];
+        pwd->h[0] = pwt->h[0];
+        break;
+    case DF_WORD:
+        pwd->w[3] = pws->w[1];
+        pwd->w[2] = pwt->w[1];
+        pwd->w[1] = pws->w[0];
+        pwd->w[0] = pwt->w[0];
+        break;
+    case DF_DOUBLE:
+        pwd->d[1] = pws->d[0];
+        pwd->d[0] = pwt->d[0];
+        break;
+    default:
+        assert(0);
+    }
+}
+
 void helper_msa_sldi_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
                         uint32_t ws, uint32_t n)
 {
-- 
2.7.4