[Beignet] [PATCH 4/7] replace rndu with llvm intrinsic.
xionghu.luo at intel.com
xionghu.luo at intel.com
Mon Mar 9 22:59:45 PDT 2015
From: Luo Xionghu <xionghu.luo at intel.com>
translate native rndu to llvm.ceil.
Signed-off-by: Luo Xionghu <xionghu.luo at intel.com>
---
backend/src/libocl/tmpl/ocl_math.tmpl.cl | 2 +-
backend/src/llvm/llvm_gen_backend.cpp | 2 --
backend/src/llvm/llvm_gen_ocl_function.hxx | 1 -
3 files changed, 1 insertion(+), 4 deletions(-)
diff --git a/backend/src/libocl/tmpl/ocl_math.tmpl.cl b/backend/src/libocl/tmpl/ocl_math.tmpl.cl
index d07e5d4..b3288b6 100644
--- a/backend/src/libocl/tmpl/ocl_math.tmpl.cl
+++ b/backend/src/libocl/tmpl/ocl_math.tmpl.cl
@@ -34,7 +34,7 @@ PURE CONST float __gen_ocl_pow(float x, float y);
PURE CONST float __gen_ocl_rcp(float x);
CONST float __gen_ocl_rndz(float x) __asm("llvm.trunc" ".f32");
CONST float __gen_ocl_rnde(float x) __asm("llvm.rint" ".f32");
-PURE CONST float __gen_ocl_rndu(float x);
+CONST float __gen_ocl_rndu(float x) __asm("llvm.ceil" ".f32");
PURE CONST float __gen_ocl_rndd(float x);
diff --git a/backend/src/llvm/llvm_gen_backend.cpp b/backend/src/llvm/llvm_gen_backend.cpp
index e358938..02d5d37 100644
--- a/backend/src/llvm/llvm_gen_backend.cpp
+++ b/backend/src/llvm/llvm_gen_backend.cpp
@@ -2674,7 +2674,6 @@ namespace gbe
case GEN_OCL_POW:
case GEN_OCL_RCP:
case GEN_OCL_ABS:
- case GEN_OCL_RNDU:
case GEN_OCL_RNDD:
case GEN_OCL_GET_IMAGE_WIDTH:
case GEN_OCL_GET_IMAGE_HEIGHT:
@@ -3076,7 +3075,6 @@ namespace gbe
}
case GEN_OCL_RSQ: this->emitUnaryCallInst(I,CS,ir::OP_RSQ); break;
case GEN_OCL_RCP: this->emitUnaryCallInst(I,CS,ir::OP_RCP); break;
- case GEN_OCL_RNDU: this->emitUnaryCallInst(I,CS,ir::OP_RNDU); break;
case GEN_OCL_RNDD: this->emitUnaryCallInst(I,CS,ir::OP_RNDD); break;
case GEN_OCL_FORCE_SIMD8: ctx.setSimdWidth(8); break;
case GEN_OCL_FORCE_SIMD16: ctx.setSimdWidth(16); break;
diff --git a/backend/src/llvm/llvm_gen_ocl_function.hxx b/backend/src/llvm/llvm_gen_ocl_function.hxx
index ea75678..83bd504 100644
--- a/backend/src/llvm/llvm_gen_ocl_function.hxx
+++ b/backend/src/llvm/llvm_gen_ocl_function.hxx
@@ -22,7 +22,6 @@ DECL_LLVM_GEN_FUNCTION(GET_WORK_DIM, __gen_ocl_get_work_dim)
DECL_LLVM_GEN_FUNCTION(RSQ, __gen_ocl_rsqrt)
DECL_LLVM_GEN_FUNCTION(POW, __gen_ocl_pow)
DECL_LLVM_GEN_FUNCTION(RCP, __gen_ocl_rcp)
-DECL_LLVM_GEN_FUNCTION(RNDU, __gen_ocl_rndu)
DECL_LLVM_GEN_FUNCTION(RNDD, __gen_ocl_rndd)
DECL_LLVM_GEN_FUNCTION(MAD, __gen_ocl_mad)
DECL_LLVM_GEN_FUNCTION(FMAX, __gen_ocl_fmax)
--
1.9.1
More information about the Beignet
mailing list