[Beignet] [PATCH 5/5] BDW: Correct stack setting in BDW.

[Yang Rong]

Remove special fftid handle for HSW in Gen8Context, and change stack buffer address
to QWORD, for curbe. Because it only waste 4 bytes register in other platform, change
to QWORD for all platform.

Signed-off-by: Yang Rong <rong.r.yang at intel.com>
---
 backend/src/backend/gen8_context.cpp | 44 ------------------------------------
 backend/src/backend/gen8_context.hpp |  2 --
 backend/src/ir/profile.cpp           |  6 ++---
 3 files changed, 3 insertions(+), 49 deletions(-)

diff --git a/backend/src/backend/gen8_context.cpp b/backend/src/backend/gen8_context.cpp
index f7484ca..cd62931 100644
--- a/backend/src/backend/gen8_context.cpp
+++ b/backend/src/backend/gen8_context.cpp
@@ -51,50 +51,6 @@ namespace gbe
     return i;
   }
 
-  void Gen8Context::emitStackPointer(void) {
-    using namespace ir;
-
-    // Only emit stack pointer computation if we use a stack
-    if (kernel->getCurbeOffset(GBE_CURBE_STACK_POINTER, 0) <= 0)
-      return;
-
-    // Check that everything is consistent in the kernel code
-    const uint32_t perLaneSize = kernel->getStackSize();
-    const uint32_t perThreadSize = perLaneSize * this->simdWidth;
-    GBE_ASSERT(perLaneSize > 0);
-    GBE_ASSERT(isPowerOf<2>(perLaneSize) == true);
-    GBE_ASSERT(isPowerOf<2>(perThreadSize) == true);
-
-    // Use shifts rather than muls which are limited to 32x16 bit sources
-    const uint32_t perLaneShift = logi2(perLaneSize);
-    const uint32_t perThreadShift = logi2(perThreadSize);
-    const GenRegister selStatckPtr = this->simdWidth == 8 ?
-      GenRegister::ud8grf(ir::ocl::stackptr) :
-      GenRegister::ud16grf(ir::ocl::stackptr);
-    const GenRegister stackptr = ra->genReg(selStatckPtr);
-    const GenRegister selStackBuffer = GenRegister::ud1grf(ir::ocl::stackbuffer);
-    const GenRegister bufferptr = ra->genReg(selStackBuffer);
-
-    // We compute the per-lane stack pointer here
-    p->push();
-      p->curr.execWidth = 1;
-      p->curr.predicate = GEN_PREDICATE_NONE;
-      //p->AND(GenRegister::ud1grf(126,0), GenRegister::ud1grf(0,5), GenRegister::immud(0x1ff));
-      p->AND(GenRegister::ud1grf(126,0), GenRegister::ud1grf(0,5), GenRegister::immud(0x7f));
-      p->AND(GenRegister::ud1grf(126,4), GenRegister::ud1grf(0,5), GenRegister::immud(0x180));
-      p->SHR(GenRegister::ud1grf(126,4), GenRegister::ud1grf(126, 4), GenRegister::immud(7));
-      p->curr.execWidth = this->simdWidth;
-      p->SHL(stackptr, stackptr, GenRegister::immud(perLaneShift));
-      p->curr.execWidth = 1;
-      p->SHL(GenRegister::ud1grf(126,0), GenRegister::ud1grf(126,0), GenRegister::immud(2));
-      p->ADD(GenRegister::ud1grf(126,0), GenRegister::ud1grf(126,0), GenRegister::ud1grf(126, 4));
-      p->SHL(GenRegister::ud1grf(126,0), GenRegister::ud1grf(126,0), GenRegister::immud(perThreadShift));
-      p->curr.execWidth = this->simdWidth;
-      p->ADD(stackptr, stackptr, bufferptr);
-      p->ADD(stackptr, stackptr, GenRegister::ud1grf(126,0));
-    p->pop();
-  }
-
   void Gen8Context::newSelection(void) {
     this->sel = GBE_NEW(Selection8, *this);
   }
diff --git a/backend/src/backend/gen8_context.hpp b/backend/src/backend/gen8_context.hpp
index 684a366..1d5b1d7
--- a/backend/src/backend/gen8_context.hpp
+++ b/backend/src/backend/gen8_context.hpp
@@ -38,8 +38,6 @@ namespace gbe
     };
     /*! device's max srcatch buffer size */
     #define GEN8_SCRATCH_SIZE  (2 * KB * KB)
-    /*! Emit the per-lane stack pointer computation */
-    virtual void emitStackPointer(void);
     /*! Align the scratch size to the device's scratch unit size */
     virtual uint32_t alignScratchSize(uint32_t size);
     /*! Get the device's max srcatch size */
diff --git a/backend/src/ir/profile.cpp b/backend/src/ir/profile.cpp
index fc69367..57cce4b
--- a/backend/src/ir/profile.cpp
+++ b/backend/src/ir/profile.cpp
@@ -48,11 +48,11 @@ namespace ir {
 
 #if GBE_DEBUG
 #define DECL_NEW_REG(FAMILY, REG, UNIFORM) \
-   r = fn.newRegister(FAMILY_DWORD, UNIFORM); \
+   r = fn.newRegister(FAMILY, UNIFORM); \
    GBE_ASSERT(r == REG);
 #else
 #define DECL_NEW_REG(FAMILY, REG, UNIFORM) \
-   fn.newRegister(FAMILY_DWORD, UNIFORM);
+   fn.newRegister(FAMILY, UNIFORM);
 #endif /* GBE_DEBUG */
     static void init(Function &fn) {
       IF_DEBUG(Register r);
@@ -75,7 +75,7 @@ namespace ir {
       DECL_NEW_REG(FAMILY_DWORD, goffset1, 1);
       DECL_NEW_REG(FAMILY_DWORD, goffset2, 1);
       DECL_NEW_REG(FAMILY_DWORD, stackptr, 0);
-      DECL_NEW_REG(FAMILY_DWORD, stackbuffer, 1);
+      DECL_NEW_REG(FAMILY_QWORD, stackbuffer, 1);
       DECL_NEW_REG(FAMILY_WORD,  blockip, 0);
       DECL_NEW_REG(FAMILY_DWORD, barrierid, 1);
       DECL_NEW_REG(FAMILY_DWORD, threadn, 1);
-- 
1.9.1