diff mbox series

[3/9] tcg: Use extract2 in tcg_gen_shifti_i64

Message ID 20190307144126.31847-4-richard.henderson@linaro.org (mailing list archive)
State New, archived
Headers show
Series tcg: Add tcg_gen_extract2_{i32,i64} | expand

Commit Message

Richard Henderson March 7, 2019, 2:41 p.m. UTC 
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tcg-op.c | 47 ++++++++++++++++++++++++-----------------------
 1 file changed, 24 insertions(+), 23 deletions(-)

Comments

Philippe Mathieu-Daudé March 9, 2019, 1 a.m. UTC | #1 
Hi Richard,

On 3/7/19 3:41 PM, Richard Henderson wrote:
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  tcg/tcg-op.c | 47 ++++++++++++++++++++++++-----------------------
>  1 file changed, 24 insertions(+), 23 deletions(-)
> 
> diff --git a/tcg/tcg-op.c b/tcg/tcg-op.c
> index deacc63e3b..34e0dbc6e0 100644
> --- a/tcg/tcg-op.c
> +++ b/tcg/tcg-op.c
> @@ -1355,31 +1355,32 @@ static inline void tcg_gen_shifti_i64(TCGv_i64 ret, TCGv_i64 arg1,
>              tcg_gen_shli_i32(TCGV_HIGH(ret), TCGV_LOW(arg1), c);
>              tcg_gen_movi_i32(TCGV_LOW(ret), 0);
>          }
> -    } else {
> -        TCGv_i32 t0, t1;
> -
> -        t0 = tcg_temp_new_i32();
> -        t1 = tcg_temp_new_i32();
> -        if (right) {
> -            tcg_gen_shli_i32(t0, TCGV_HIGH(arg1), 32 - c);
> -            if (arith) {
> -                tcg_gen_sari_i32(t1, TCGV_HIGH(arg1), c);
> -            } else {
> -                tcg_gen_shri_i32(t1, TCGV_HIGH(arg1), c);
> -            }
> -            tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
> -            tcg_gen_or_i32(TCGV_LOW(ret), TCGV_LOW(ret), t0);
> -            tcg_gen_mov_i32(TCGV_HIGH(ret), t1);
> +    } else if (right) {
> +        if (TCG_TARGET_HAS_extract2_i32) {
> +            tcg_gen_extract2_i32(TCGV_LOW(ret), TCGV_LOW(arg1),
> +                                 TCGV_HIGH(arg1), c);
>          } else {
> -            tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
> -            /* Note: ret can be the same as arg1, so we use t1 */
> -            tcg_gen_shli_i32(t1, TCGV_LOW(arg1), c);
> -            tcg_gen_shli_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
> -            tcg_gen_or_i32(TCGV_HIGH(ret), TCGV_HIGH(ret), t0);
> -            tcg_gen_mov_i32(TCGV_LOW(ret), t1);
> +            tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
> +            tcg_gen_deposit_i32(TCGV_LOW(ret), TCGV_LOW(ret),
> +                                TCGV_HIGH(arg1), 32 - c, c);
>          }
> -        tcg_temp_free_i32(t0);
> -        tcg_temp_free_i32(t1);
> +        if (arith) {
> +            tcg_gen_sari_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
> +        } else {
> +            tcg_gen_shri_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
> +        }
> +    } else {
> +        if (TCG_TARGET_HAS_extract2_i32) {
> +            tcg_gen_extract2_i32(TCGV_HIGH(ret), TCGV_LOW(arg1),
> +                                 TCGV_HIGH(arg1), 32 - c);
> +        } else {
> +            TCGv_i32 t0 = tcg_temp_new_i32();
> +            tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
> +            tcg_gen_deposit_i32(TCGV_HIGH(ret), t0,
> +                                TCGV_HIGH(arg1), c, 32 - c);
> +            tcg_temp_free_i32(t0);
> +        }
> +        tcg_gen_shli_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
>      }
>  }

I find this patch quite hard to review because you somehow do more than
simply use extract2, you also reordered part of this function.

I find it easier to review as split in 2 commits:

1/ reorder; the code movement is way easier to follow:

-- >8 --
@@ -1355,31 +1355,22 @@ static inline void tcg_gen_shifti_i64(TCGv_i64
ret, TCGv_i64 arg1,
             tcg_gen_shli_i32(TCGV_HIGH(ret), TCGV_LOW(arg1), c);
             tcg_gen_movi_i32(TCGV_LOW(ret), 0);
         }
-    } else {
-        TCGv_i32 t0, t1;
-
-        t0 = tcg_temp_new_i32();
-        t1 = tcg_temp_new_i32();
-        if (right) {
-            tcg_gen_shli_i32(t0, TCGV_HIGH(arg1), 32 - c);
-            if (arith) {
-                tcg_gen_sari_i32(t1, TCGV_HIGH(arg1), c);
-            } else {
-                tcg_gen_shri_i32(t1, TCGV_HIGH(arg1), c);
-            }
-            tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
-            tcg_gen_or_i32(TCGV_LOW(ret), TCGV_LOW(ret), t0);
-            tcg_gen_mov_i32(TCGV_HIGH(ret), t1);
+    } else if (right) {
+        tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
+        tcg_gen_deposit_i32(TCGV_LOW(ret), TCGV_LOW(ret),
+                            TCGV_HIGH(arg1), 32 - c, c);
+        if (arith) {
+            tcg_gen_sari_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
         } else {
-            tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
-            /* Note: ret can be the same as arg1, so we use t1 */
-            tcg_gen_shli_i32(t1, TCGV_LOW(arg1), c);
-            tcg_gen_shli_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
-            tcg_gen_or_i32(TCGV_HIGH(ret), TCGV_HIGH(ret), t0);
-            tcg_gen_mov_i32(TCGV_LOW(ret), t1);
+            tcg_gen_shri_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
         }
+    } else {
+        TCGv_i32 t0 = tcg_temp_new_i32();
+        tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
+        tcg_gen_deposit_i32(TCGV_HIGH(ret), t0,
+                            TCGV_HIGH(arg1), c, 32 - c);
         tcg_temp_free_i32(t0);
-        tcg_temp_free_i32(t1);
+        tcg_gen_shli_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
     }
 }
---

2/ use extract2:

-- >8 --
@@ -1356,20 +1356,30 @@ static inline void tcg_gen_shifti_i64(TCGv_i64
ret, TCGv_i64 arg1,
             tcg_gen_movi_i32(TCGV_LOW(ret), 0);
         }
     } else if (right) {
-        tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
-        tcg_gen_deposit_i32(TCGV_LOW(ret), TCGV_LOW(ret),
-                            TCGV_HIGH(arg1), 32 - c, c);
+        if (TCG_TARGET_HAS_extract2_i32) {
+            tcg_gen_extract2_i32(TCGV_LOW(ret),
+                                 TCGV_LOW(arg1), TCGV_HIGH(arg1), c);
+        } else {
+            tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
+            tcg_gen_deposit_i32(TCGV_LOW(ret), TCGV_LOW(ret),
+                                TCGV_HIGH(arg1), 32 - c, c);
+        }
         if (arith) {
             tcg_gen_sari_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
         } else {
             tcg_gen_shri_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
         }
     } else {
-        TCGv_i32 t0 = tcg_temp_new_i32();
-        tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
-        tcg_gen_deposit_i32(TCGV_HIGH(ret), t0,
-                            TCGV_HIGH(arg1), c, 32 - c);
-        tcg_temp_free_i32(t0);
+        if (TCG_TARGET_HAS_extract2_i32) {
+            tcg_gen_extract2_i32(TCGV_HIGH(ret),
+                                 TCGV_LOW(arg1), TCGV_HIGH(arg1), 32 - c);
+        } else {
+            TCGv_i32 t0 = tcg_temp_new_i32();
+            tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
+            tcg_gen_deposit_i32(TCGV_HIGH(ret), t0,
+                                TCGV_HIGH(arg1), c, 32 - c);
+            tcg_temp_free_i32(t0);
+        }
         tcg_gen_shli_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
     }
 }
---

Note that I indented the extract2 slighly differently,
TCGV_LOW(arg1) and TCGV_HIGH(arg1) are now together on the same line.
My brain find it more digestible :)

Regards,

Phil.
Aleksandar Markovic March 9, 2019, 8:30 p.m. UTC | #2 
On Thursday, March 7, 2019, Richard Henderson <richard.henderson@linaro.org>
wrote:

> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  tcg/tcg-op.c | 47 ++++++++++++++++++++++++-----------------------
>  1 file changed, 24 insertions(+), 23 deletions(-)
>
>
Extract2 is not a good name for this new function, IMHO.



> diff --git a/tcg/tcg-op.c b/tcg/tcg-op.c
> index deacc63e3b..34e0dbc6e0 100644
> --- a/tcg/tcg-op.c
> +++ b/tcg/tcg-op.c
> @@ -1355,31 +1355,32 @@ static inline void tcg_gen_shifti_i64(TCGv_i64
> ret, TCGv_i64 arg1,
>              tcg_gen_shli_i32(TCGV_HIGH(ret), TCGV_LOW(arg1), c);
>              tcg_gen_movi_i32(TCGV_LOW(ret), 0);
>          }
> -    } else {
> -        TCGv_i32 t0, t1;
> -
> -        t0 = tcg_temp_new_i32();
> -        t1 = tcg_temp_new_i32();
> -        if (right) {
> -            tcg_gen_shli_i32(t0, TCGV_HIGH(arg1), 32 - c);
> -            if (arith) {
> -                tcg_gen_sari_i32(t1, TCGV_HIGH(arg1), c);
> -            } else {
> -                tcg_gen_shri_i32(t1, TCGV_HIGH(arg1), c);
> -            }
> -            tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
> -            tcg_gen_or_i32(TCGV_LOW(ret), TCGV_LOW(ret), t0);
> -            tcg_gen_mov_i32(TCGV_HIGH(ret), t1);
> +    } else if (right) {
> +        if (TCG_TARGET_HAS_extract2_i32) {
> +            tcg_gen_extract2_i32(TCGV_LOW(ret), TCGV_LOW(arg1),
> +                                 TCGV_HIGH(arg1), c);
>          } else {
> -            tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
> -            /* Note: ret can be the same as arg1, so we use t1 */
> -            tcg_gen_shli_i32(t1, TCGV_LOW(arg1), c);
> -            tcg_gen_shli_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
> -            tcg_gen_or_i32(TCGV_HIGH(ret), TCGV_HIGH(ret), t0);
> -            tcg_gen_mov_i32(TCGV_LOW(ret), t1);
> +            tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
> +            tcg_gen_deposit_i32(TCGV_LOW(ret), TCGV_LOW(ret),
> +                                TCGV_HIGH(arg1), 32 - c, c);
>          }
> -        tcg_temp_free_i32(t0);
> -        tcg_temp_free_i32(t1);
> +        if (arith) {
> +            tcg_gen_sari_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
> +        } else {
> +            tcg_gen_shri_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
> +        }
> +    } else {
> +        if (TCG_TARGET_HAS_extract2_i32) {
> +            tcg_gen_extract2_i32(TCGV_HIGH(ret), TCGV_LOW(arg1),
> +                                 TCGV_HIGH(arg1), 32 - c);
> +        } else {
> +            TCGv_i32 t0 = tcg_temp_new_i32();
> +            tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
> +            tcg_gen_deposit_i32(TCGV_HIGH(ret), t0,
> +                                TCGV_HIGH(arg1), c, 32 - c);
> +            tcg_temp_free_i32(t0);
> +        }
> +        tcg_gen_shli_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
>      }
>  }
>
> --
> 2.17.2
>
>
>
Richard Henderson March 10, 2019, 6:43 a.m. UTC | #3 
On 3/9/19 12:30 PM, Aleksandar Markovic wrote:
> 
> 
> On Thursday, March 7, 2019, Richard Henderson <richard.henderson@linaro.org
> <mailto:richard.henderson@linaro.org>> wrote:
> 
>     Signed-off-by: Richard Henderson <richard.henderson@linaro.org
>     <mailto:richard.henderson@linaro.org>>
>     ---
>      tcg/tcg-op.c | 47 ++++++++++++++++++++++++-----------------------
>      1 file changed, 24 insertions(+), 23 deletions(-)
> 
> 
> Extract2 is not a good name for this new function, IMHO.

If you're going to bike shed the name, you should suggest something else.

But I think it's a pretty good name, since it extracts one register output out
of two register inputs.


r~
diff mbox series

Patch

diff --git a/tcg/tcg-op.c b/tcg/tcg-op.c
index deacc63e3b..34e0dbc6e0 100644
--- a/tcg/tcg-op.c
+++ b/tcg/tcg-op.c
@@ -1355,31 +1355,32 @@  static inline void tcg_gen_shifti_i64(TCGv_i64 ret, TCGv_i64 arg1,
             tcg_gen_shli_i32(TCGV_HIGH(ret), TCGV_LOW(arg1), c);
             tcg_gen_movi_i32(TCGV_LOW(ret), 0);
         }
-    } else {
-        TCGv_i32 t0, t1;
-
-        t0 = tcg_temp_new_i32();
-        t1 = tcg_temp_new_i32();
-        if (right) {
-            tcg_gen_shli_i32(t0, TCGV_HIGH(arg1), 32 - c);
-            if (arith) {
-                tcg_gen_sari_i32(t1, TCGV_HIGH(arg1), c);
-            } else {
-                tcg_gen_shri_i32(t1, TCGV_HIGH(arg1), c);
-            }
-            tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
-            tcg_gen_or_i32(TCGV_LOW(ret), TCGV_LOW(ret), t0);
-            tcg_gen_mov_i32(TCGV_HIGH(ret), t1);
+    } else if (right) {
+        if (TCG_TARGET_HAS_extract2_i32) {
+            tcg_gen_extract2_i32(TCGV_LOW(ret), TCGV_LOW(arg1),
+                                 TCGV_HIGH(arg1), c);
         } else {
-            tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
-            /* Note: ret can be the same as arg1, so we use t1 */
-            tcg_gen_shli_i32(t1, TCGV_LOW(arg1), c);
-            tcg_gen_shli_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
-            tcg_gen_or_i32(TCGV_HIGH(ret), TCGV_HIGH(ret), t0);
-            tcg_gen_mov_i32(TCGV_LOW(ret), t1);
+            tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
+            tcg_gen_deposit_i32(TCGV_LOW(ret), TCGV_LOW(ret),
+                                TCGV_HIGH(arg1), 32 - c, c);
         }
-        tcg_temp_free_i32(t0);
-        tcg_temp_free_i32(t1);
+        if (arith) {
+            tcg_gen_sari_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
+        } else {
+            tcg_gen_shri_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
+        }
+    } else {
+        if (TCG_TARGET_HAS_extract2_i32) {
+            tcg_gen_extract2_i32(TCGV_HIGH(ret), TCGV_LOW(arg1),
+                                 TCGV_HIGH(arg1), 32 - c);
+        } else {
+            TCGv_i32 t0 = tcg_temp_new_i32();
+            tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
+            tcg_gen_deposit_i32(TCGV_HIGH(ret), t0,
+                                TCGV_HIGH(arg1), c, 32 - c);
+            tcg_temp_free_i32(t0);
+        }
+        tcg_gen_shli_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
     }
 }