drm/radeon: fixup tiling group size and backendmap on r6xx-r9xx (v4)
Tiling group size is always 256bits on r6xx/r7xx/r8xx/9xx. Also fix and
simplify render backend map. This now properly sets up the backend map
on r6xx-9xx which should improve 3D performance.
Vadim benchmarked also:
Some benchmarks on juniper (5750), fullscreen 1920x1080,
first result - kernel 3.4.0+ (fb21affa), second - with these patches:
Lightsmark: 91 fps => 123 fps +35%
Doom3: 74 fps => 101 fps +36%
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c
index f388a1d..45cfcea 100644
--- a/drivers/gpu/drm/radeon/r600.c
+++ b/drivers/gpu/drm/radeon/r600.c
@@ -1376,113 +1376,51 @@
return r600_gpu_soft_reset(rdev);
}
-static u32 r600_get_tile_pipe_to_backend_map(u32 num_tile_pipes,
- u32 num_backends,
- u32 backend_disable_mask)
+u32 r6xx_remap_render_backend(struct radeon_device *rdev,
+ u32 tiling_pipe_num,
+ u32 max_rb_num,
+ u32 total_max_rb_num,
+ u32 disabled_rb_mask)
{
- u32 backend_map = 0;
- u32 enabled_backends_mask;
- u32 enabled_backends_count;
- u32 cur_pipe;
- u32 swizzle_pipe[R6XX_MAX_PIPES];
- u32 cur_backend;
- u32 i;
+ u32 rendering_pipe_num, rb_num_width, req_rb_num;
+ u32 pipe_rb_ratio, pipe_rb_remain;
+ u32 data = 0, mask = 1 << (max_rb_num - 1);
+ unsigned i, j;
- if (num_tile_pipes > R6XX_MAX_PIPES)
- num_tile_pipes = R6XX_MAX_PIPES;
- if (num_tile_pipes < 1)
- num_tile_pipes = 1;
- if (num_backends > R6XX_MAX_BACKENDS)
- num_backends = R6XX_MAX_BACKENDS;
- if (num_backends < 1)
- num_backends = 1;
+ /* mask out the RBs that don't exist on that asic */
+ disabled_rb_mask |= (0xff << max_rb_num) & 0xff;
- enabled_backends_mask = 0;
- enabled_backends_count = 0;
- for (i = 0; i < R6XX_MAX_BACKENDS; ++i) {
- if (((backend_disable_mask >> i) & 1) == 0) {
- enabled_backends_mask |= (1 << i);
- ++enabled_backends_count;
+ rendering_pipe_num = 1 << tiling_pipe_num;
+ req_rb_num = total_max_rb_num - r600_count_pipe_bits(disabled_rb_mask);
+ BUG_ON(rendering_pipe_num < req_rb_num);
+
+ pipe_rb_ratio = rendering_pipe_num / req_rb_num;
+ pipe_rb_remain = rendering_pipe_num - pipe_rb_ratio * req_rb_num;
+
+ if (rdev->family <= CHIP_RV740) {
+ /* r6xx/r7xx */
+ rb_num_width = 2;
+ } else {
+ /* eg+ */
+ rb_num_width = 4;
+ }
+
+ for (i = 0; i < max_rb_num; i++) {
+ if (!(mask & disabled_rb_mask)) {
+ for (j = 0; j < pipe_rb_ratio; j++) {
+ data <<= rb_num_width;
+ data |= max_rb_num - i - 1;
+ }
+ if (pipe_rb_remain) {
+ data <<= rb_num_width;
+ data |= max_rb_num - i - 1;
+ pipe_rb_remain--;
+ }
}
- if (enabled_backends_count == num_backends)
- break;
+ mask >>= 1;
}
- if (enabled_backends_count == 0) {
- enabled_backends_mask = 1;
- enabled_backends_count = 1;
- }
-
- if (enabled_backends_count != num_backends)
- num_backends = enabled_backends_count;
-
- memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * R6XX_MAX_PIPES);
- switch (num_tile_pipes) {
- case 1:
- swizzle_pipe[0] = 0;
- break;
- case 2:
- swizzle_pipe[0] = 0;
- swizzle_pipe[1] = 1;
- break;
- case 3:
- swizzle_pipe[0] = 0;
- swizzle_pipe[1] = 1;
- swizzle_pipe[2] = 2;
- break;
- case 4:
- swizzle_pipe[0] = 0;
- swizzle_pipe[1] = 1;
- swizzle_pipe[2] = 2;
- swizzle_pipe[3] = 3;
- break;
- case 5:
- swizzle_pipe[0] = 0;
- swizzle_pipe[1] = 1;
- swizzle_pipe[2] = 2;
- swizzle_pipe[3] = 3;
- swizzle_pipe[4] = 4;
- break;
- case 6:
- swizzle_pipe[0] = 0;
- swizzle_pipe[1] = 2;
- swizzle_pipe[2] = 4;
- swizzle_pipe[3] = 5;
- swizzle_pipe[4] = 1;
- swizzle_pipe[5] = 3;
- break;
- case 7:
- swizzle_pipe[0] = 0;
- swizzle_pipe[1] = 2;
- swizzle_pipe[2] = 4;
- swizzle_pipe[3] = 6;
- swizzle_pipe[4] = 1;
- swizzle_pipe[5] = 3;
- swizzle_pipe[6] = 5;
- break;
- case 8:
- swizzle_pipe[0] = 0;
- swizzle_pipe[1] = 2;
- swizzle_pipe[2] = 4;
- swizzle_pipe[3] = 6;
- swizzle_pipe[4] = 1;
- swizzle_pipe[5] = 3;
- swizzle_pipe[6] = 5;
- swizzle_pipe[7] = 7;
- break;
- }
-
- cur_backend = 0;
- for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
- while (((1 << cur_backend) & enabled_backends_mask) == 0)
- cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
-
- backend_map |= (u32)(((cur_backend & 3) << (swizzle_pipe[cur_pipe] * 2)));
-
- cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
- }
-
- return backend_map;
+ return data;
}
int r600_count_pipe_bits(uint32_t val)
@@ -1500,7 +1438,6 @@
{
u32 tiling_config;
u32 ramcfg;
- u32 backend_map;
u32 cc_rb_backend_disable;
u32 cc_gc_shader_pipe_config;
u32 tmp;
@@ -1511,8 +1448,9 @@
u32 sq_thread_resource_mgmt = 0;
u32 sq_stack_resource_mgmt_1 = 0;
u32 sq_stack_resource_mgmt_2 = 0;
+ u32 disabled_rb_mask;
- /* FIXME: implement */
+ rdev->config.r600.tiling_group_size = 256;
switch (rdev->family) {
case CHIP_R600:
rdev->config.r600.max_pipes = 4;
@@ -1616,10 +1554,7 @@
rdev->config.r600.tiling_nbanks = 4 << ((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT);
tiling_config |= BANK_TILING((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT);
tiling_config |= GROUP_SIZE((ramcfg & BURSTLENGTH_MASK) >> BURSTLENGTH_SHIFT);
- if ((ramcfg & BURSTLENGTH_MASK) >> BURSTLENGTH_SHIFT)
- rdev->config.r600.tiling_group_size = 512;
- else
- rdev->config.r600.tiling_group_size = 256;
+
tmp = (ramcfg & NOOFROWS_MASK) >> NOOFROWS_SHIFT;
if (tmp > 3) {
tiling_config |= ROW_TILING(3);
@@ -1631,32 +1566,36 @@
tiling_config |= BANK_SWAPS(1);
cc_rb_backend_disable = RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000;
- cc_rb_backend_disable |=
- BACKEND_DISABLE((R6XX_MAX_BACKENDS_MASK << rdev->config.r600.max_backends) & R6XX_MAX_BACKENDS_MASK);
+ tmp = R6XX_MAX_BACKENDS -
+ r600_count_pipe_bits((cc_rb_backend_disable >> 16) & R6XX_MAX_BACKENDS_MASK);
+ if (tmp < rdev->config.r600.max_backends) {
+ rdev->config.r600.max_backends = tmp;
+ }
- cc_gc_shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG) & 0xffffff00;
- cc_gc_shader_pipe_config |=
- INACTIVE_QD_PIPES((R6XX_MAX_PIPES_MASK << rdev->config.r600.max_pipes) & R6XX_MAX_PIPES_MASK);
- cc_gc_shader_pipe_config |=
- INACTIVE_SIMDS((R6XX_MAX_SIMDS_MASK << rdev->config.r600.max_simds) & R6XX_MAX_SIMDS_MASK);
+ cc_gc_shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG) & 0x00ffff00;
+ tmp = R6XX_MAX_PIPES -
+ r600_count_pipe_bits((cc_gc_shader_pipe_config >> 8) & R6XX_MAX_PIPES_MASK);
+ if (tmp < rdev->config.r600.max_pipes) {
+ rdev->config.r600.max_pipes = tmp;
+ }
+ tmp = R6XX_MAX_SIMDS -
+ r600_count_pipe_bits((cc_gc_shader_pipe_config >> 16) & R6XX_MAX_SIMDS_MASK);
+ if (tmp < rdev->config.r600.max_simds) {
+ rdev->config.r600.max_simds = tmp;
+ }
- backend_map = r600_get_tile_pipe_to_backend_map(rdev->config.r600.max_tile_pipes,
- (R6XX_MAX_BACKENDS -
- r600_count_pipe_bits((cc_rb_backend_disable &
- R6XX_MAX_BACKENDS_MASK) >> 16)),
- (cc_rb_backend_disable >> 16));
+ disabled_rb_mask = (RREG32(CC_RB_BACKEND_DISABLE) >> 16) & R6XX_MAX_BACKENDS_MASK;
+ tmp = (tiling_config & PIPE_TILING__MASK) >> PIPE_TILING__SHIFT;
+ tmp = r6xx_remap_render_backend(rdev, tmp, rdev->config.r600.max_backends,
+ R6XX_MAX_BACKENDS, disabled_rb_mask);
+ tiling_config |= tmp << 16;
+ rdev->config.r600.backend_map = tmp;
+
rdev->config.r600.tile_config = tiling_config;
- rdev->config.r600.backend_map = backend_map;
- tiling_config |= BACKEND_MAP(backend_map);
WREG32(GB_TILING_CONFIG, tiling_config);
WREG32(DCP_TILING_CONFIG, tiling_config & 0xffff);
WREG32(HDP_TILING_CONFIG, tiling_config & 0xffff);
- /* Setup pipes */
- WREG32(CC_RB_BACKEND_DISABLE, cc_rb_backend_disable);
- WREG32(CC_GC_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
- WREG32(GC_USER_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
-
tmp = R6XX_MAX_PIPES - r600_count_pipe_bits((cc_gc_shader_pipe_config & INACTIVE_QD_PIPES_MASK) >> 8);
WREG32(VGT_OUT_DEALLOC_CNTL, (tmp * 4) & DEALLOC_DIST_MASK);
WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, ((tmp * 4) - 2) & VTX_REUSE_DEPTH_MASK);