drm/amdgpu: optimize amdgpu_vm_update_ptes a bit
authorChristian König <christian.koenig@amd.com>
Tue, 26 Jan 2016 11:37:49 +0000 (12:37 +0100)
committerAlex Deucher <alexander.deucher@amd.com>
Wed, 10 Feb 2016 19:17:07 +0000 (14:17 -0500)
Don't calculate the end address multiple times.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c

index 0aa43c5..eb38aa2 100644 (file)
@@ -568,6 +568,10 @@ static void amdgpu_vm_frag_ptes(struct amdgpu_device *adev,
 
        unsigned count;
 
+       /* Abort early if there isn't anything to do */
+       if (pe_start == pe_end)
+               return;
+
        /* system pages are non continuously */
        if (gtt || !(flags & AMDGPU_PTE_VALID) || (frag_start >= frag_end)) {
 
@@ -622,9 +626,9 @@ static void amdgpu_vm_update_ptes(struct amdgpu_device *adev,
                                  uint64_t start, uint64_t end,
                                  uint64_t dst, uint32_t flags)
 {
-       uint64_t mask = AMDGPU_VM_PTE_COUNT - 1;
-       uint64_t last_pte = ~0, last_dst = ~0;
-       unsigned count = 0;
+       const uint64_t mask = AMDGPU_VM_PTE_COUNT - 1;
+
+       uint64_t last_pe_start = ~0, last_pe_end = ~0, last_dst = ~0;
        uint64_t addr;
 
        /* walk over the address space and update the page tables */
@@ -632,40 +636,36 @@ static void amdgpu_vm_update_ptes(struct amdgpu_device *adev,
                uint64_t pt_idx = addr >> amdgpu_vm_block_size;
                struct amdgpu_bo *pt = vm->page_tables[pt_idx].entry.robj;
                unsigned nptes;
-               uint64_t pte;
+               uint64_t pe_start;
 
                if ((addr & ~mask) == (end & ~mask))
                        nptes = end - addr;
                else
                        nptes = AMDGPU_VM_PTE_COUNT - (addr & mask);
 
-               pte = amdgpu_bo_gpu_offset(pt);
-               pte += (addr & mask) * 8;
+               pe_start = amdgpu_bo_gpu_offset(pt);
+               pe_start += (addr & mask) * 8;
 
-               if ((last_pte + 8 * count) != pte) {
+               if (last_pe_end != pe_start) {
 
-                       if (count) {
-                               amdgpu_vm_frag_ptes(adev, gtt, gtt_flags, ib,
-                                                   last_pte, last_pte + 8 * count,
-                                                   last_dst, flags);
-                       }
+                       amdgpu_vm_frag_ptes(adev, gtt, gtt_flags, ib,
+                                           last_pe_start, last_pe_end,
+                                           last_dst, flags);
 
-                       count = nptes;
-                       last_pte = pte;
+                       last_pe_start = pe_start;
+                       last_pe_end = pe_start + 8 * nptes;
                        last_dst = dst;
                } else {
-                       count += nptes;
+                       last_pe_end += 8 * nptes;
                }
 
                addr += nptes;
                dst += nptes * AMDGPU_GPU_PAGE_SIZE;
        }
 
-       if (count) {
-               amdgpu_vm_frag_ptes(adev, gtt, gtt_flags, ib,
-                                   last_pte, last_pte + 8 * count,
-                                   last_dst, flags);
-       }
+       amdgpu_vm_frag_ptes(adev, gtt, gtt_flags, ib,
+                           last_pe_start, last_pe_end,
+                           last_dst, flags);
 }
 
 /**