Fixed SNB build

diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index 067e99e..d7a10e3 100644 (file)
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -64,6 +64,7 @@ ADD_EXECUTABLE(dct tests/dct.c)
 ADD_EXECUTABLE(binomialOption tests/binomialOption.c)
 ADD_EXECUTABLE(nbody tests/nbody.c)
 ADD_EXECUTABLE(svm_test tests/svm_test.c)
+ADD_EXECUTABLE(fast_mat_mul tests/fast_mat_mul.cpp)
 TARGET_LINK_LIBRARIES(test_copy_buffer cl_test m)
 TARGET_LINK_LIBRARIES(test_copy_image cl_test m)
 TARGET_LINK_LIBRARIES(test_enqueue_read cl_test m)
@@ -88,4 +89,5 @@ TARGET_LINK_LIBRARIES(dct cl_test m)
 TARGET_LINK_LIBRARIES(binomialOption cl_test m)
 TARGET_LINK_LIBRARIES(nbody cl_test m)
 TARGET_LINK_LIBRARIES(svm_test cl_test m)
+TARGET_LINK_LIBRARIES(fast_mat_mul cl_test m)
 

diff --git a/src/cl_command_queue_gen7.c b/src/cl_command_queue_gen7.c
index 8c93b73..dce1c2c 100644 (file)
--- a/src/cl_command_queue_gen7.c
+++ b/src/cl_command_queue_gen7.c
@@ -118,6 +118,7 @@ cl_command_queue_ND_range_gen7(cl_command_queue queue,
 
   /* Check that the local work sizes are OK */
   TRY (cl_kernel_work_group_sz, ker, local_wk_sz, 3, &local_sz);
+  //kernel.thread_n = thread_n = local_sz / simd_sz;
   kernel.thread_n = thread_n = local_sz / simd_sz;
 
   /* CURBE step 1. Allocate and fill fields shared by threads in workgroup */

diff --git a/src/cl_kernel.h b/src/cl_kernel.h
index d11d551..c27cff6 100644 (file)
--- a/src/cl_kernel.h
+++ b/src/cl_kernel.h
@@ -288,7 +288,11 @@ struct _cl_kernel {
 };
 
 /* Size of the surface state as encoded in the binary blob */
+#if USE_OLD_COMPILER
+#define SURFACE_SZ 32
+#else
 #define SURFACE_SZ 64
+#endif
 
 /* Allocate an empty kernel */
 extern cl_kernel cl_kernel_new(void);

diff --git a/src/intel/intel_gpgpu.c b/src/intel/intel_gpgpu.c
index 753a9bd..10e20db 100644 (file)
--- a/src/intel/intel_gpgpu.c
+++ b/src/intel/intel_gpgpu.c
@@ -177,10 +177,19 @@ gpgpu_load_vfe_state(intel_gpgpu_t *state)
   vfe->vfe1.reset_gateway_timer = 1;
   vfe->vfe1.max_threads = state->max_threads - 1;
   vfe->vfe1.urb_entries = 64;
-  vfe->vfe3.curbe_size = 63;
-  vfe->vfe3.urbe_size = 13;
-  vfe->vfe4.scoreboard_mask =
-    (state->drv->gen_ver == 7 || state->drv->gen_ver == 75) ? 0 : 0x80000000;
+  if (state->drv->gen_ver >= 7) {
+    vfe->vfe3.curbe_size = 480;
+    vfe->vfe4.scoreboard_mask = 0;
+  } else {
+    vfe->vfe3.curbe_size = 63;
+    vfe->vfe3.urb_size = 13;
+    vfe->vfe4.scoreboard_mask = 0x80000000;
+  }
+
+  //M
+  //p
+  //vfe->vfe3.urb_size = 13;
+  //vfe->vfe4.scoreboard_mask = (state->drv->gen_ver == 7 || state->drv->gen_ver == 75) ? 0 : 0x80000000;
   intel_batchbuffer_alloc_space(state->batch, sizeof(gen6_vfe_state_inline_t));
   ADVANCE_BATCH(state->batch);
 }
@@ -191,9 +200,14 @@ gpgpu_load_constant_buffer(intel_gpgpu_t *state)
   BEGIN_BATCH(state->batch, 4);
   OUT_BATCH(state->batch, CMD(2,0,1) | (4 - 2));  /* length-2 */
   OUT_BATCH(state->batch, 0);                     /* mbz */
+// XXX
+#if 1
   OUT_BATCH(state->batch,
             state->urb.size_cs_entry*
             state->urb.num_cs_entries*32);
+#else
+  OUT_BATCH(state->batch, 5120);
+#endif
   OUT_RELOC(state->batch, state->curbe_b.bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
   ADVANCE_BATCH(state->batch);
 }
@@ -204,7 +218,7 @@ gpgpu_load_idrt(intel_gpgpu_t *state)
   BEGIN_BATCH(state->batch, 4);
   OUT_BATCH(state->batch, CMD(2,0,2) | (4 - 2)); /* length-2 */
   OUT_BATCH(state->batch, 0);                    /* mbz */
-  OUT_BATCH(state->batch, state->idrt_b.num*32);
+  OUT_BATCH(state->batch, state->idrt_b.num << 5);
   OUT_RELOC(state->batch, state->idrt_b.bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
   ADVANCE_BATCH(state->batch);
 }

diff --git a/src/intel/intel_structs.h b/src/intel/intel_structs.h
index 0301692..84b5a17 100644 (file)
--- a/src/intel/intel_structs.h
+++ b/src/intel/intel_structs.h
@@ -234,7 +234,7 @@ typedef struct gen6_vfe_state_inline
 
   struct {
     uint32_t curbe_size:16; /* in GRFs */
-    uint32_t urbe_size:16;  /* in GRFs */
+    uint32_t urb_size:16;  /* in GRFs */
   } vfe3;
 
   struct {