[Beignet] [PATCH] GBE/libocl: Fix sub_sat corner case.

Ruiling Song ruiling.song at intel.com
Sun Sep 14 20:14:05 PDT 2014


It seems that hw return wrong result when y is equal to 0x80000000
in sub_sat(int x, int y). So we re-write it as:
add_sat(add_sat(0x7fffffff, x), 1)

Also enable corresponding utest.

Signed-off-by: Ruiling Song <ruiling.song at intel.com>
---
 backend/src/libocl/tmpl/ocl_integer.tmpl.cl |    2 +-
 utests/compiler_saturate_sub.cpp            |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/backend/src/libocl/tmpl/ocl_integer.tmpl.cl b/backend/src/libocl/tmpl/ocl_integer.tmpl.cl
index d33d70e..9230604 100644
--- a/backend/src/libocl/tmpl/ocl_integer.tmpl.cl
+++ b/backend/src/libocl/tmpl/ocl_integer.tmpl.cl
@@ -100,7 +100,7 @@ OVERLOADABLE int ocl_sadd_sat(int x, int y);
 OVERLOADABLE int add_sat(int x, int y) { return ocl_sadd_sat(x, y); }
 OVERLOADABLE int ocl_ssub_sat(int x, int y);
 OVERLOADABLE int sub_sat(int x, int y) {
-  return (y == 0x80000000u) ? (x & 0x7FFFFFFF) : ocl_ssub_sat(x, y);
+  return (y == 0x80000000u) ? (ocl_sadd_sat(ocl_sadd_sat(0x7fffffff, x), 1)) : ocl_ssub_sat(x, y);
 }
 OVERLOADABLE long ocl_sadd_sat(long x, long y);
 OVERLOADABLE long add_sat(long x, long y) {
diff --git a/utests/compiler_saturate_sub.cpp b/utests/compiler_saturate_sub.cpp
index 48947b7..1c95e2d 100644
--- a/utests/compiler_saturate_sub.cpp
+++ b/utests/compiler_saturate_sub.cpp
@@ -108,7 +108,7 @@ compiler_saturate_sub(int8_t, test_char)
 compiler_saturate_sub(uint8_t, test_uchar)
 compiler_saturate_sub(int16_t, test_short)
 compiler_saturate_sub(uint16_t, test_ushort)
-//compiler_saturate_sub(int32_t, test_int) // TODO due to the possible hardware bug, we disable this, uncomment it when it's done.
+compiler_saturate_sub(int32_t, test_int)
 compiler_saturate_sub(uint32_t, test_uint)
 //compiler_saturate_sub(int64_t, test_long)
 //compiler_saturate_sub(uint64_t, test_ulong)
-- 
1.7.10.4



More information about the Beignet mailing list