libocl: Consider only bottom ilogb(2m-1)+1 bits

Submitted by Song, Ruiling on Sept. 20, 2017, 8:43 a.m.

Details

Message ID 148B1B7A67D1C24B9EF0BE42EA49770684E160EF@SHSMSX103.ccr.corp.intel.com
State New
Series "libocl: Consider only bottom ilogb(2m-1)+1 bits"
Headers show

Commit Message

Song, Ruiling Sept. 20, 2017, 8:43 a.m.
LGTM

Ruiling

-----Original Message-----
From: Beignet [mailto:beignet-bounces@lists.freedesktop.org] On Behalf Of Jan Vesely

Sent: Tuesday, September 19, 2017 1:13 PM
To: beignet@lists.freedesktop.org
Subject: [Beignet] [PATCH] libocl: Consider only bottom ilogb(2m-1)+1 bits

Signed-off-by: Jan Vesely <jano.vesely@gmail.com>

---
Fixes remaining shuffle2 piglit tests on Skylake
 backend/src/libocl/src/ocl_misc.cl | 60 +++++++++++++++++++-------------------
 1 file changed, 30 insertions(+), 30 deletions(-)

-- 
2.13.5

_______________________________________________
Beignet mailing list
Beignet@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/beignet

Patch hide | download patch | download mbox

diff --git a/backend/src/libocl/src/ocl_misc.cl b/backend/src/libocl/src/ocl_misc.cl
index d8e09aed..f104f4ff 100644
--- a/backend/src/libocl/src/ocl_misc.cl
+++ b/backend/src/libocl/src/ocl_misc.cl
@@ -109,8 +109,8 @@  DEF(double)
 #define DEC2X(TYPE, MASKTYPE) \
   OVERLOADABLE TYPE##2 shuffle2(TYPE##16 x, TYPE##16 y, MASKTYPE##2 mask) { \
     TYPE##2 z; \
-    z.s0 = mask.s0 < 16 ? ((TYPE *)&x)[mask.s0] : ((TYPE *)&y)[mask.s0 & 15]; \
-    z.s1 = mask.s1 < 16 ? ((TYPE *)&x)[mask.s1] : ((TYPE *)&y)[mask.s1 & 15]; \
+    z.s0 = (mask.s0 & 31) < 16 ? ((TYPE *)&x)[mask.s0 & 31] : ((TYPE *)&y)[mask.s0 & 15]; \
+    z.s1 = (mask.s1 & 31) < 16 ? ((TYPE *)&x)[mask.s1 & 31] : ((TYPE *)&y)[mask.s1 & 15]; \
     return z; \
   }
 
@@ -122,10 +122,10 @@  DEF(double)
 #define DEC4X(TYPE, MASKTYPE) \
   OVERLOADABLE TYPE##4 shuffle2(TYPE##16 x, TYPE##16 y, MASKTYPE##4 mask) { \
     TYPE##4 z; \
-    z.s0 = mask.s0 < 16 ? ((TYPE *)&x)[mask.s0] : ((TYPE *)&y)[mask.s0 & 15]; \
-    z.s1 = mask.s1 < 16 ? ((TYPE *)&x)[mask.s1] : ((TYPE *)&y)[mask.s1 & 15]; \
-    z.s2 = mask.s2 < 16 ? ((TYPE *)&x)[mask.s2] : ((TYPE *)&y)[mask.s2 & 15]; \
-    z.s3 = mask.s3 < 16 ? ((TYPE *)&x)[mask.s3] : ((TYPE *)&y)[mask.s3 & 15]; \
+    z.s0 = (mask.s0 & 31) < 16 ? ((TYPE *)&x)[mask.s0 & 31] : ((TYPE *)&y)[mask.s0 & 15]; \
+    z.s1 = (mask.s1 & 31) < 16 ? ((TYPE *)&x)[mask.s1 & 31] : ((TYPE *)&y)[mask.s1 & 15]; \
+    z.s2 = (mask.s2 & 31) < 16 ? ((TYPE *)&x)[mask.s2 & 31] : ((TYPE *)&y)[mask.s2 & 15]; \
+    z.s3 = (mask.s3 & 31) < 16 ? ((TYPE *)&x)[mask.s3 & 31] : ((TYPE *)&y)[mask.s3 & 15]; \
     return z; \
   }
 
@@ -137,14 +137,14 @@  DEF(double)
 #define DEC8X(TYPE, MASKTYPE) \
   OVERLOADABLE TYPE##8 shuffle2(TYPE##16 x, TYPE##16 y, MASKTYPE##8 mask) { \
     TYPE##8 z; \
-    z.s0 = mask.s0 < 16 ? ((TYPE *)&x)[mask.s0] : ((TYPE *)&y)[mask.s0 & 15]; \
-    z.s1 = mask.s1 < 16 ? ((TYPE *)&x)[mask.s1] : ((TYPE *)&y)[mask.s1 & 15]; \
-    z.s2 = mask.s2 < 16 ? ((TYPE *)&x)[mask.s2] : ((TYPE *)&y)[mask.s2 & 15]; \
-    z.s3 = mask.s3 < 16 ? ((TYPE *)&x)[mask.s3] : ((TYPE *)&y)[mask.s3 & 15]; \
-    z.s4 = mask.s4 < 16 ? ((TYPE *)&x)[mask.s4] : ((TYPE *)&y)[mask.s4 & 15]; \
-    z.s5 = mask.s5 < 16 ? ((TYPE *)&x)[mask.s5] : ((TYPE *)&y)[mask.s5 & 15]; \
-    z.s6 = mask.s6 < 16 ? ((TYPE *)&x)[mask.s6] : ((TYPE *)&y)[mask.s6 & 15]; \
-    z.s7 = mask.s7 < 16 ? ((TYPE *)&x)[mask.s7] : ((TYPE *)&y)[mask.s7 & 15]; \
+    z.s0 = (mask.s0 & 31) < 16 ? ((TYPE *)&x)[mask.s0 & 31] : ((TYPE *)&y)[mask.s0 & 15]; \
+    z.s1 = (mask.s1 & 31) < 16 ? ((TYPE *)&x)[mask.s1 & 31] : ((TYPE *)&y)[mask.s1 & 15]; \
+    z.s2 = (mask.s2 & 31) < 16 ? ((TYPE *)&x)[mask.s2 & 31] : ((TYPE *)&y)[mask.s2 & 15]; \
+    z.s3 = (mask.s3 & 31) < 16 ? ((TYPE *)&x)[mask.s3 & 31] : ((TYPE *)&y)[mask.s3 & 15]; \
+    z.s4 = (mask.s4 & 31) < 16 ? ((TYPE *)&x)[mask.s4 & 31] : ((TYPE *)&y)[mask.s4 & 15]; \
+    z.s5 = (mask.s5 & 31) < 16 ? ((TYPE *)&x)[mask.s5 & 31] : ((TYPE *)&y)[mask.s5 & 15]; \
+    z.s6 = (mask.s6 & 31) < 16 ? ((TYPE *)&x)[mask.s6 & 31] : ((TYPE *)&y)[mask.s6 & 15]; \
+    z.s7 = (mask.s7 & 31) < 16 ? ((TYPE *)&x)[mask.s7 & 31] : ((TYPE *)&y)[mask.s7 & 15]; \
     return z; \
   }
 
@@ -156,22 +156,22 @@  DEF(double)
 #define DEC16X(TYPE, MASKTYPE) \
   OVERLOADABLE TYPE##16 shuffle2(TYPE##16 x, TYPE##16 y, MASKTYPE##16 mask) { \
     TYPE##16 z; \
-    z.s0 = mask.s0 < 16 ? ((TYPE *)&x)[mask.s0] : ((TYPE *)&y)[mask.s0 & 15]; \
-    z.s1 = mask.s1 < 16 ? ((TYPE *)&x)[mask.s1] : ((TYPE *)&y)[mask.s1 & 15]; \
-    z.s2 = mask.s2 < 16 ? ((TYPE *)&x)[mask.s2] : ((TYPE *)&y)[mask.s2 & 15]; \
-    z.s3 = mask.s3 < 16 ? ((TYPE *)&x)[mask.s3] : ((TYPE *)&y)[mask.s3 & 15]; \
-    z.s4 = mask.s4 < 16 ? ((TYPE *)&x)[mask.s4] : ((TYPE *)&y)[mask.s4 & 15]; \
-    z.s5 = mask.s5 < 16 ? ((TYPE *)&x)[mask.s5] : ((TYPE *)&y)[mask.s5 & 15]; \
-    z.s6 = mask.s6 < 16 ? ((TYPE *)&x)[mask.s6] : ((TYPE *)&y)[mask.s6 & 15]; \
-    z.s7 = mask.s7 < 16 ? ((TYPE *)&x)[mask.s7] : ((TYPE *)&y)[mask.s7 & 15]; \
-    z.s8 = mask.s8 < 16 ? ((TYPE *)&x)[mask.s8] : ((TYPE *)&y)[mask.s8 & 15]; \
-    z.s9 = mask.s9 < 16 ? ((TYPE *)&x)[mask.s9] : ((TYPE *)&y)[mask.s9 & 15]; \
-    z.sA = mask.sA < 16 ? ((TYPE *)&x)[mask.sA] : ((TYPE *)&y)[mask.sA & 15]; \
-    z.sB = mask.sB < 16 ? ((TYPE *)&x)[mask.sB] : ((TYPE *)&y)[mask.sB & 15]; \
-    z.sC = mask.sC < 16 ? ((TYPE *)&x)[mask.sC] : ((TYPE *)&y)[mask.sC & 15]; \
-    z.sD = mask.sD < 16 ? ((TYPE *)&x)[mask.sD] : ((TYPE *)&y)[mask.sD & 15]; \
-    z.sE = mask.sE < 16 ? ((TYPE *)&x)[mask.sE] : ((TYPE *)&y)[mask.sE & 15]; \
-    z.sF = mask.sF < 16 ? ((TYPE *)&x)[mask.sF] : ((TYPE *)&y)[mask.sF & 15]; \
+    z.s0 = (mask.s0 & 31) < 16 ? ((TYPE *)&x)[mask.s0 & 31] : ((TYPE *)&y)[mask.s0 & 15]; \
+    z.s1 = (mask.s1 & 31) < 16 ? ((TYPE *)&x)[mask.s1 & 31] : ((TYPE *)&y)[mask.s1 & 15]; \
+    z.s2 = (mask.s2 & 31) < 16 ? ((TYPE *)&x)[mask.s2 & 31] : ((TYPE *)&y)[mask.s2 & 15]; \
+    z.s3 = (mask.s3 & 31) < 16 ? ((TYPE *)&x)[mask.s3 & 31] : ((TYPE *)&y)[mask.s3 & 15]; \
+    z.s4 = (mask.s4 & 31) < 16 ? ((TYPE *)&x)[mask.s4 & 31] : ((TYPE *)&y)[mask.s4 & 15]; \
+    z.s5 = (mask.s5 & 31) < 16 ? ((TYPE *)&x)[mask.s5 & 31] : ((TYPE *)&y)[mask.s5 & 15]; \
+    z.s6 = (mask.s6 & 31) < 16 ? ((TYPE *)&x)[mask.s6 & 31] : ((TYPE *)&y)[mask.s6 & 15]; \
+    z.s7 = (mask.s7 & 31) < 16 ? ((TYPE *)&x)[mask.s7 & 31] : ((TYPE *)&y)[mask.s7 & 15]; \
+    z.s8 = (mask.s8 & 31) < 16 ? ((TYPE *)&x)[mask.s8 & 31] : ((TYPE *)&y)[mask.s8 & 15]; \
+    z.s9 = (mask.s9 & 31) < 16 ? ((TYPE *)&x)[mask.s9 & 31] : ((TYPE *)&y)[mask.s9 & 15]; \
+    z.sA = (mask.sA & 31) < 16 ? ((TYPE *)&x)[mask.sA & 31] : ((TYPE *)&y)[mask.sA & 15]; \
+    z.sB = (mask.sB & 31) < 16 ? ((TYPE *)&x)[mask.sB & 31] : ((TYPE *)&y)[mask.sB & 15]; \
+    z.sC = (mask.sC & 31) < 16 ? ((TYPE *)&x)[mask.sC & 31] : ((TYPE *)&y)[mask.sC & 15]; \
+    z.sD = (mask.sD & 31) < 16 ? ((TYPE *)&x)[mask.sD & 31] : ((TYPE *)&y)[mask.sD & 15]; \
+    z.sE = (mask.sE & 31) < 16 ? ((TYPE *)&x)[mask.sE & 31] : ((TYPE *)&y)[mask.sE & 15]; \
+    z.sF = (mask.sF & 31) < 16 ? ((TYPE *)&x)[mask.sF & 31] : ((TYPE *)&y)[mask.sF & 15]; \
     return z; \
   }
 

Comments

Yang, Rong R Sept. 21, 2017, 7:14 a.m.
Pushed, thanks.

> -----Original Message-----

> From: Beignet [mailto:beignet-bounces@lists.freedesktop.org] On Behalf Of

> Song, Ruiling

> Sent: Wednesday, September 20, 2017 4:44 PM

> To: Jan Vesely <jano.vesely@gmail.com>; beignet@lists.freedesktop.org

> Subject: Re: [Beignet] [PATCH] libocl: Consider only bottom ilogb(2m-1)+1 bits

> 

> LGTM

> 

> Ruiling

> 

> -----Original Message-----

> From: Beignet [mailto:beignet-bounces@lists.freedesktop.org] On Behalf Of Jan

> Vesely

> Sent: Tuesday, September 19, 2017 1:13 PM

> To: beignet@lists.freedesktop.org

> Subject: [Beignet] [PATCH] libocl: Consider only bottom ilogb(2m-1)+1 bits

> 

> Signed-off-by: Jan Vesely <jano.vesely@gmail.com>

> ---

> Fixes remaining shuffle2 piglit tests on Skylake

> backend/src/libocl/src/ocl_misc.cl | 60 +++++++++++++++++++-------------------

>  1 file changed, 30 insertions(+), 30 deletions(-)

> 

> diff --git a/backend/src/libocl/src/ocl_misc.cl

> b/backend/src/libocl/src/ocl_misc.cl

> index d8e09aed..f104f4ff 100644

> --- a/backend/src/libocl/src/ocl_misc.cl

> +++ b/backend/src/libocl/src/ocl_misc.cl

> @@ -109,8 +109,8 @@ DEF(double)

>  #define DEC2X(TYPE, MASKTYPE) \

>    OVERLOADABLE TYPE##2 shuffle2(TYPE##16 x, TYPE##16 y, MASKTYPE##2

> mask) { \

>      TYPE##2 z; \

> -    z.s0 = mask.s0 < 16 ? ((TYPE *)&x)[mask.s0] : ((TYPE *)&y)[mask.s0 & 15]; \

> -    z.s1 = mask.s1 < 16 ? ((TYPE *)&x)[mask.s1] : ((TYPE *)&y)[mask.s1 & 15]; \

> +    z.s0 = (mask.s0 & 31) < 16 ? ((TYPE *)&x)[mask.s0 & 31] : ((TYPE

> *)&y)[mask.s0 & 15]; \

> +    z.s1 = (mask.s1 & 31) < 16 ? ((TYPE *)&x)[mask.s1 & 31] : ((TYPE

> + *)&y)[mask.s1 & 15]; \

>      return z; \

>    }

> 

> @@ -122,10 +122,10 @@ DEF(double)

>  #define DEC4X(TYPE, MASKTYPE) \

>    OVERLOADABLE TYPE##4 shuffle2(TYPE##16 x, TYPE##16 y, MASKTYPE##4

> mask) { \

>      TYPE##4 z; \

> -    z.s0 = mask.s0 < 16 ? ((TYPE *)&x)[mask.s0] : ((TYPE *)&y)[mask.s0 & 15]; \

> -    z.s1 = mask.s1 < 16 ? ((TYPE *)&x)[mask.s1] : ((TYPE *)&y)[mask.s1 & 15]; \

> -    z.s2 = mask.s2 < 16 ? ((TYPE *)&x)[mask.s2] : ((TYPE *)&y)[mask.s2 & 15]; \

> -    z.s3 = mask.s3 < 16 ? ((TYPE *)&x)[mask.s3] : ((TYPE *)&y)[mask.s3 & 15]; \

> +    z.s0 = (mask.s0 & 31) < 16 ? ((TYPE *)&x)[mask.s0 & 31] : ((TYPE

> *)&y)[mask.s0 & 15]; \

> +    z.s1 = (mask.s1 & 31) < 16 ? ((TYPE *)&x)[mask.s1 & 31] : ((TYPE

> *)&y)[mask.s1 & 15]; \

> +    z.s2 = (mask.s2 & 31) < 16 ? ((TYPE *)&x)[mask.s2 & 31] : ((TYPE

> *)&y)[mask.s2 & 15]; \

> +    z.s3 = (mask.s3 & 31) < 16 ? ((TYPE *)&x)[mask.s3 & 31] : ((TYPE

> + *)&y)[mask.s3 & 15]; \

>      return z; \

>    }

> 

> @@ -137,14 +137,14 @@ DEF(double)

>  #define DEC8X(TYPE, MASKTYPE) \

>    OVERLOADABLE TYPE##8 shuffle2(TYPE##16 x, TYPE##16 y, MASKTYPE##8

> mask) { \

>      TYPE##8 z; \

> -    z.s0 = mask.s0 < 16 ? ((TYPE *)&x)[mask.s0] : ((TYPE *)&y)[mask.s0 & 15]; \

> -    z.s1 = mask.s1 < 16 ? ((TYPE *)&x)[mask.s1] : ((TYPE *)&y)[mask.s1 & 15]; \

> -    z.s2 = mask.s2 < 16 ? ((TYPE *)&x)[mask.s2] : ((TYPE *)&y)[mask.s2 & 15]; \

> -    z.s3 = mask.s3 < 16 ? ((TYPE *)&x)[mask.s3] : ((TYPE *)&y)[mask.s3 & 15]; \

> -    z.s4 = mask.s4 < 16 ? ((TYPE *)&x)[mask.s4] : ((TYPE *)&y)[mask.s4 & 15]; \

> -    z.s5 = mask.s5 < 16 ? ((TYPE *)&x)[mask.s5] : ((TYPE *)&y)[mask.s5 & 15]; \

> -    z.s6 = mask.s6 < 16 ? ((TYPE *)&x)[mask.s6] : ((TYPE *)&y)[mask.s6 & 15]; \

> -    z.s7 = mask.s7 < 16 ? ((TYPE *)&x)[mask.s7] : ((TYPE *)&y)[mask.s7 & 15]; \

> +    z.s0 = (mask.s0 & 31) < 16 ? ((TYPE *)&x)[mask.s0 & 31] : ((TYPE

> *)&y)[mask.s0 & 15]; \

> +    z.s1 = (mask.s1 & 31) < 16 ? ((TYPE *)&x)[mask.s1 & 31] : ((TYPE

> *)&y)[mask.s1 & 15]; \

> +    z.s2 = (mask.s2 & 31) < 16 ? ((TYPE *)&x)[mask.s2 & 31] : ((TYPE

> *)&y)[mask.s2 & 15]; \

> +    z.s3 = (mask.s3 & 31) < 16 ? ((TYPE *)&x)[mask.s3 & 31] : ((TYPE

> *)&y)[mask.s3 & 15]; \

> +    z.s4 = (mask.s4 & 31) < 16 ? ((TYPE *)&x)[mask.s4 & 31] : ((TYPE

> *)&y)[mask.s4 & 15]; \

> +    z.s5 = (mask.s5 & 31) < 16 ? ((TYPE *)&x)[mask.s5 & 31] : ((TYPE

> *)&y)[mask.s5 & 15]; \

> +    z.s6 = (mask.s6 & 31) < 16 ? ((TYPE *)&x)[mask.s6 & 31] : ((TYPE

> *)&y)[mask.s6 & 15]; \

> +    z.s7 = (mask.s7 & 31) < 16 ? ((TYPE *)&x)[mask.s7 & 31] : ((TYPE

> + *)&y)[mask.s7 & 15]; \

>      return z; \

>    }

> 

> @@ -156,22 +156,22 @@ DEF(double)

>  #define DEC16X(TYPE, MASKTYPE) \

>    OVERLOADABLE TYPE##16 shuffle2(TYPE##16 x, TYPE##16 y, MASKTYPE##16

> mask) { \

>      TYPE##16 z; \

> -    z.s0 = mask.s0 < 16 ? ((TYPE *)&x)[mask.s0] : ((TYPE *)&y)[mask.s0 & 15]; \

> -    z.s1 = mask.s1 < 16 ? ((TYPE *)&x)[mask.s1] : ((TYPE *)&y)[mask.s1 & 15]; \

> -    z.s2 = mask.s2 < 16 ? ((TYPE *)&x)[mask.s2] : ((TYPE *)&y)[mask.s2 & 15]; \

> -    z.s3 = mask.s3 < 16 ? ((TYPE *)&x)[mask.s3] : ((TYPE *)&y)[mask.s3 & 15]; \

> -    z.s4 = mask.s4 < 16 ? ((TYPE *)&x)[mask.s4] : ((TYPE *)&y)[mask.s4 & 15]; \

> -    z.s5 = mask.s5 < 16 ? ((TYPE *)&x)[mask.s5] : ((TYPE *)&y)[mask.s5 & 15]; \

> -    z.s6 = mask.s6 < 16 ? ((TYPE *)&x)[mask.s6] : ((TYPE *)&y)[mask.s6 & 15]; \

> -    z.s7 = mask.s7 < 16 ? ((TYPE *)&x)[mask.s7] : ((TYPE *)&y)[mask.s7 & 15]; \

> -    z.s8 = mask.s8 < 16 ? ((TYPE *)&x)[mask.s8] : ((TYPE *)&y)[mask.s8 & 15]; \

> -    z.s9 = mask.s9 < 16 ? ((TYPE *)&x)[mask.s9] : ((TYPE *)&y)[mask.s9 & 15]; \

> -    z.sA = mask.sA < 16 ? ((TYPE *)&x)[mask.sA] : ((TYPE *)&y)[mask.sA & 15]; \

> -    z.sB = mask.sB < 16 ? ((TYPE *)&x)[mask.sB] : ((TYPE *)&y)[mask.sB & 15]; \

> -    z.sC = mask.sC < 16 ? ((TYPE *)&x)[mask.sC] : ((TYPE *)&y)[mask.sC & 15]; \

> -    z.sD = mask.sD < 16 ? ((TYPE *)&x)[mask.sD] : ((TYPE *)&y)[mask.sD & 15]; \

> -    z.sE = mask.sE < 16 ? ((TYPE *)&x)[mask.sE] : ((TYPE *)&y)[mask.sE & 15]; \

> -    z.sF = mask.sF < 16 ? ((TYPE *)&x)[mask.sF] : ((TYPE *)&y)[mask.sF & 15]; \

> +    z.s0 = (mask.s0 & 31) < 16 ? ((TYPE *)&x)[mask.s0 & 31] : ((TYPE

> *)&y)[mask.s0 & 15]; \

> +    z.s1 = (mask.s1 & 31) < 16 ? ((TYPE *)&x)[mask.s1 & 31] : ((TYPE

> *)&y)[mask.s1 & 15]; \

> +    z.s2 = (mask.s2 & 31) < 16 ? ((TYPE *)&x)[mask.s2 & 31] : ((TYPE

> *)&y)[mask.s2 & 15]; \

> +    z.s3 = (mask.s3 & 31) < 16 ? ((TYPE *)&x)[mask.s3 & 31] : ((TYPE

> *)&y)[mask.s3 & 15]; \

> +    z.s4 = (mask.s4 & 31) < 16 ? ((TYPE *)&x)[mask.s4 & 31] : ((TYPE

> *)&y)[mask.s4 & 15]; \

> +    z.s5 = (mask.s5 & 31) < 16 ? ((TYPE *)&x)[mask.s5 & 31] : ((TYPE

> *)&y)[mask.s5 & 15]; \

> +    z.s6 = (mask.s6 & 31) < 16 ? ((TYPE *)&x)[mask.s6 & 31] : ((TYPE

> *)&y)[mask.s6 & 15]; \

> +    z.s7 = (mask.s7 & 31) < 16 ? ((TYPE *)&x)[mask.s7 & 31] : ((TYPE

> *)&y)[mask.s7 & 15]; \

> +    z.s8 = (mask.s8 & 31) < 16 ? ((TYPE *)&x)[mask.s8 & 31] : ((TYPE

> *)&y)[mask.s8 & 15]; \

> +    z.s9 = (mask.s9 & 31) < 16 ? ((TYPE *)&x)[mask.s9 & 31] : ((TYPE

> *)&y)[mask.s9 & 15]; \

> +    z.sA = (mask.sA & 31) < 16 ? ((TYPE *)&x)[mask.sA & 31] : ((TYPE

> *)&y)[mask.sA & 15]; \

> +    z.sB = (mask.sB & 31) < 16 ? ((TYPE *)&x)[mask.sB & 31] : ((TYPE

> *)&y)[mask.sB & 15]; \

> +    z.sC = (mask.sC & 31) < 16 ? ((TYPE *)&x)[mask.sC & 31] : ((TYPE

> *)&y)[mask.sC & 15]; \

> +    z.sD = (mask.sD & 31) < 16 ? ((TYPE *)&x)[mask.sD & 31] : ((TYPE

> *)&y)[mask.sD & 15]; \

> +    z.sE = (mask.sE & 31) < 16 ? ((TYPE *)&x)[mask.sE & 31] : ((TYPE

> *)&y)[mask.sE & 15]; \

> +    z.sF = (mask.sF & 31) < 16 ? ((TYPE *)&x)[mask.sF & 31] : ((TYPE

> + *)&y)[mask.sF & 15]; \

>      return z; \

>    }

> 

> --

> 2.13.5

> 

> _______________________________________________

> Beignet mailing list

> Beignet@lists.freedesktop.org

> https://lists.freedesktop.org/mailman/listinfo/beignet

> _______________________________________________

> Beignet mailing list

> Beignet@lists.freedesktop.org

> https://lists.freedesktop.org/mailman/listinfo/beignet