[Beignet] [PATCH 5/7] replace rndd with llvm intrinsic.

[xionghu.luo at intel.com]

From: Luo Xionghu <xionghu.luo at intel.com>

translate native rndd to llvm.floor.

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      | 4 ++--
 backend/src/llvm/llvm_gen_ocl_function.hxx | 1 -
 3 files changed, 3 insertions(+), 4 deletions(-)

diff --git a/backend/src/libocl/tmpl/ocl_math.tmpl.cl b/backend/src/libocl/tmpl/ocl_math.tmpl.cl
index b3288b6..40b6401 100644
--- a/backend/src/libocl/tmpl/ocl_math.tmpl.cl
+++ b/backend/src/libocl/tmpl/ocl_math.tmpl.cl
@@ -35,7 +35,7 @@ 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");
 CONST float __gen_ocl_rndu(float x) __asm("llvm.ceil" ".f32");
-PURE CONST float __gen_ocl_rndd(float x);
+CONST float __gen_ocl_rndd(float x) __asm("llvm.floor" ".f32");
 
 
 /* native functions */
diff --git a/backend/src/llvm/llvm_gen_backend.cpp b/backend/src/llvm/llvm_gen_backend.cpp
index 02d5d37..6549950 100644
--- a/backend/src/llvm/llvm_gen_backend.cpp
+++ b/backend/src/llvm/llvm_gen_backend.cpp
@@ -2613,6 +2613,7 @@ namespace gbe
           case Intrinsic::fma:
           case Intrinsic::trunc:
           case Intrinsic::rint:
+          case Intrinsic::floor:
           case Intrinsic::sin:
           case Intrinsic::cos:
           case Intrinsic::log2:
@@ -2674,7 +2675,6 @@ namespace gbe
       case GEN_OCL_POW:
       case GEN_OCL_RCP:
       case GEN_OCL_ABS:
-      case GEN_OCL_RNDD:
       case GEN_OCL_GET_IMAGE_WIDTH:
       case GEN_OCL_GET_IMAGE_HEIGHT:
       case GEN_OCL_GET_IMAGE_CHANNEL_DATA_TYPE:
@@ -2999,6 +2999,7 @@ namespace gbe
           case Intrinsic::fabs: this->emitUnaryCallInst(I,CS,ir::OP_ABS); break;
           case Intrinsic::trunc: this->emitUnaryCallInst(I,CS,ir::OP_RNDZ); break;
           case Intrinsic::rint: this->emitUnaryCallInst(I,CS,ir::OP_RNDE); break;
+          case Intrinsic::floor: this->emitUnaryCallInst(I,CS,ir::OP_RNDD); break;
           case Intrinsic::sin: this->emitUnaryCallInst(I,CS,ir::OP_SIN); break;
           case Intrinsic::cos: this->emitUnaryCallInst(I,CS,ir::OP_COS); break;
           case Intrinsic::log2: this->emitUnaryCallInst(I,CS,ir::OP_LOG); break;
@@ -3075,7 +3076,6 @@ 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_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;
           case GEN_OCL_LBARRIER: ctx.SYNC(ir::syncLocalBarrier); break;
diff --git a/backend/src/llvm/llvm_gen_ocl_function.hxx b/backend/src/llvm/llvm_gen_ocl_function.hxx
index 83bd504..2cc63bd 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(RNDD, __gen_ocl_rndd)
 DECL_LLVM_GEN_FUNCTION(MAD, __gen_ocl_mad)
 DECL_LLVM_GEN_FUNCTION(FMAX, __gen_ocl_fmax)
 DECL_LLVM_GEN_FUNCTION(FMIN, __gen_ocl_fmin)
-- 
1.9.1