Commit 416a2bd2 authored by Alex Deucher's avatar Alex Deucher Committed by Dave Airlie
Browse files

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: default avatarAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: default avatarJerome Glisse <jglisse@redhat.com>
Signed-off-by: default avatarDave Airlie <airlied@redhat.com>
parent 95c4b23e

drivers/gpu/drm/radeon/evergreen.c

+45 −323
Original line number Diff line number Diff line
@@ -1558,163 +1558,10 @@ int evergreen_cp_resume(struct radeon_device *rdev) 
/*

 * Core functions

 */

static u32 evergreen_get_tile_pipe_to_backend_map(struct radeon_device *rdev,

						  u32 num_tile_pipes,

						  u32 num_backends,

						  u32 backend_disable_mask)

{

	u32 backend_map = 0;

	u32 enabled_backends_mask = 0;

	u32 enabled_backends_count = 0;

	u32 cur_pipe;

	u32 swizzle_pipe[EVERGREEN_MAX_PIPES];

	u32 cur_backend = 0;

	u32 i;

	bool force_no_swizzle;



	if (num_tile_pipes > EVERGREEN_MAX_PIPES)

		num_tile_pipes = EVERGREEN_MAX_PIPES;

	if (num_tile_pipes < 1)

		num_tile_pipes = 1;

	if (num_backends > EVERGREEN_MAX_BACKENDS)

		num_backends = EVERGREEN_MAX_BACKENDS;

	if (num_backends < 1)

		num_backends = 1;



	for (i = 0; i < EVERGREEN_MAX_BACKENDS; ++i) {

		if (((backend_disable_mask >> i) & 1) == 0) {

			enabled_backends_mask |= (1 << i);

			++enabled_backends_count;

		}

		if (enabled_backends_count == num_backends)

			break;

	}



	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) * EVERGREEN_MAX_PIPES);

	switch (rdev->family) {

	case CHIP_CEDAR:

	case CHIP_REDWOOD:

	case CHIP_PALM:

	case CHIP_SUMO:

	case CHIP_SUMO2:

	case CHIP_TURKS:

	case CHIP_CAICOS:

		force_no_swizzle = false;

		break;

	case CHIP_CYPRESS:

	case CHIP_HEMLOCK:

	case CHIP_JUNIPER:

	case CHIP_BARTS:

	default:

		force_no_swizzle = true;

		break;

	}

	if (force_no_swizzle) {

		bool last_backend_enabled = false;



		force_no_swizzle = false;

		for (i = 0; i < EVERGREEN_MAX_BACKENDS; ++i) {

			if (((enabled_backends_mask >> i) & 1) == 1) {

				if (last_backend_enabled)

					force_no_swizzle = true;

				last_backend_enabled = true;

			} else

				last_backend_enabled = false;

		}

	}



	switch (num_tile_pipes) {

	case 1:

	case 3:

	case 5:

	case 7:

		DRM_ERROR("odd number of pipes!\n");

		break;

	case 2:

		swizzle_pipe[0] = 0;

		swizzle_pipe[1] = 1;

		break;

	case 4:

		if (force_no_swizzle) {

			swizzle_pipe[0] = 0;

			swizzle_pipe[1] = 1;

			swizzle_pipe[2] = 2;

			swizzle_pipe[3] = 3;

		} else {

			swizzle_pipe[0] = 0;

			swizzle_pipe[1] = 2;

			swizzle_pipe[2] = 1;

			swizzle_pipe[3] = 3;

		}

		break;

	case 6:

		if (force_no_swizzle) {

			swizzle_pipe[0] = 0;

			swizzle_pipe[1] = 1;

			swizzle_pipe[2] = 2;

			swizzle_pipe[3] = 3;

			swizzle_pipe[4] = 4;

			swizzle_pipe[5] = 5;

		} else {

			swizzle_pipe[0] = 0;

			swizzle_pipe[1] = 2;

			swizzle_pipe[2] = 4;

			swizzle_pipe[3] = 1;

			swizzle_pipe[4] = 3;

			swizzle_pipe[5] = 5;

		}

		break;

	case 8:

		if (force_no_swizzle) {

			swizzle_pipe[0] = 0;

			swizzle_pipe[1] = 1;

			swizzle_pipe[2] = 2;

			swizzle_pipe[3] = 3;

			swizzle_pipe[4] = 4;

			swizzle_pipe[5] = 5;

			swizzle_pipe[6] = 6;

			swizzle_pipe[7] = 7;

		} else {

			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;

	}



	for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {

		while (((1 << cur_backend) & enabled_backends_mask) == 0)

			cur_backend = (cur_backend + 1) % EVERGREEN_MAX_BACKENDS;



		backend_map |= (((cur_backend & 0xf) << (swizzle_pipe[cur_pipe] * 4)));



		cur_backend = (cur_backend + 1) % EVERGREEN_MAX_BACKENDS;

	}



	return backend_map;

}



static void evergreen_gpu_init(struct radeon_device *rdev)

{

	u32 cc_rb_backend_disable = 0;

	u32 cc_gc_shader_pipe_config;

	u32 gb_addr_config = 0;

	u32 gb_addr_config;

	u32 mc_shared_chmap, mc_arb_ramcfg;

	u32 gb_backend_map;

	u32 grbm_gfx_index;

	u32 sx_debug_1;

	u32 smx_dc_ctl0;

	u32 sq_config;
@@ -1729,6 +1576,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
	u32 sq_stack_resource_mgmt_3;

	u32 vgt_cache_invalidation;

	u32 hdp_host_path_cntl, tmp;

	u32 disabled_rb_mask;

	int i, j, num_shader_engines, ps_thread_count;



	switch (rdev->family) {
@@ -1753,6 +1601,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		rdev->config.evergreen.sc_prim_fifo_size = 0x100;

		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;

		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;

		gb_addr_config = CYPRESS_GB_ADDR_CONFIG_GOLDEN;

		break;

	case CHIP_JUNIPER:

		rdev->config.evergreen.num_ses = 1;
@@ -1774,6 +1623,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		rdev->config.evergreen.sc_prim_fifo_size = 0x100;

		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;

		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;

		gb_addr_config = JUNIPER_GB_ADDR_CONFIG_GOLDEN;

		break;

	case CHIP_REDWOOD:

		rdev->config.evergreen.num_ses = 1;
@@ -1795,6 +1645,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		rdev->config.evergreen.sc_prim_fifo_size = 0x100;

		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;

		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;

		gb_addr_config = REDWOOD_GB_ADDR_CONFIG_GOLDEN;

		break;

	case CHIP_CEDAR:

	default:
@@ -1817,6 +1668,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		rdev->config.evergreen.sc_prim_fifo_size = 0x40;

		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;

		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;

		gb_addr_config = CEDAR_GB_ADDR_CONFIG_GOLDEN;

		break;

	case CHIP_PALM:

		rdev->config.evergreen.num_ses = 1;
@@ -1838,6 +1690,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		rdev->config.evergreen.sc_prim_fifo_size = 0x40;

		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;

		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;

		gb_addr_config = CEDAR_GB_ADDR_CONFIG_GOLDEN;

		break;

	case CHIP_SUMO:

		rdev->config.evergreen.num_ses = 1;
@@ -1865,6 +1718,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		rdev->config.evergreen.sc_prim_fifo_size = 0x40;

		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;

		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;

		gb_addr_config = REDWOOD_GB_ADDR_CONFIG_GOLDEN;

		break;

	case CHIP_SUMO2:

		rdev->config.evergreen.num_ses = 1;
@@ -1886,6 +1740,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		rdev->config.evergreen.sc_prim_fifo_size = 0x40;

		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;

		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;

		gb_addr_config = REDWOOD_GB_ADDR_CONFIG_GOLDEN;

		break;

	case CHIP_BARTS:

		rdev->config.evergreen.num_ses = 2;
@@ -1907,6 +1762,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		rdev->config.evergreen.sc_prim_fifo_size = 0x100;

		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;

		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;

		gb_addr_config = BARTS_GB_ADDR_CONFIG_GOLDEN;

		break;

	case CHIP_TURKS:

		rdev->config.evergreen.num_ses = 1;
@@ -1928,6 +1784,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		rdev->config.evergreen.sc_prim_fifo_size = 0x100;

		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;

		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;

		gb_addr_config = TURKS_GB_ADDR_CONFIG_GOLDEN;

		break;

	case CHIP_CAICOS:

		rdev->config.evergreen.num_ses = 1;
@@ -1949,6 +1806,7 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		rdev->config.evergreen.sc_prim_fifo_size = 0x40;

		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;

		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;

		gb_addr_config = CAICOS_GB_ADDR_CONFIG_GOLDEN;

		break;

	}
@@ -1965,20 +1823,6 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 


	evergreen_fix_pci_max_read_req_size(rdev);



	cc_gc_shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG) & ~2;



	cc_gc_shader_pipe_config |=

		INACTIVE_QD_PIPES((EVERGREEN_MAX_PIPES_MASK << rdev->config.evergreen.max_pipes)

				  & EVERGREEN_MAX_PIPES_MASK);

	cc_gc_shader_pipe_config |=

		INACTIVE_SIMDS((EVERGREEN_MAX_SIMDS_MASK << rdev->config.evergreen.max_simds)

			       & EVERGREEN_MAX_SIMDS_MASK);



	cc_rb_backend_disable =

		BACKEND_DISABLE((EVERGREEN_MAX_BACKENDS_MASK << rdev->config.evergreen.max_backends)

				& EVERGREEN_MAX_BACKENDS_MASK);





	mc_shared_chmap = RREG32(MC_SHARED_CHMAP);

	if ((rdev->family == CHIP_PALM) ||

	    (rdev->family == CHIP_SUMO) ||
@@ -1987,134 +1831,6 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
	else

		mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);



	switch (rdev->config.evergreen.max_tile_pipes) {

	case 1:

	default:

		gb_addr_config |= NUM_PIPES(0);

		break;

	case 2:

		gb_addr_config |= NUM_PIPES(1);

		break;

	case 4:

		gb_addr_config |= NUM_PIPES(2);

		break;

	case 8:

		gb_addr_config |= NUM_PIPES(3);

		break;

	}



	gb_addr_config |= PIPE_INTERLEAVE_SIZE((mc_arb_ramcfg & BURSTLENGTH_MASK) >> BURSTLENGTH_SHIFT);

	gb_addr_config |= BANK_INTERLEAVE_SIZE(0);

	gb_addr_config |= NUM_SHADER_ENGINES(rdev->config.evergreen.num_ses - 1);

	gb_addr_config |= SHADER_ENGINE_TILE_SIZE(1);

	gb_addr_config |= NUM_GPUS(0); /* Hemlock? */

	gb_addr_config |= MULTI_GPU_TILE_SIZE(2);



	if (((mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT) > 2)

		gb_addr_config |= ROW_SIZE(2);

	else

		gb_addr_config |= ROW_SIZE((mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT);



	if (rdev->ddev->pdev->device == 0x689e) {

		u32 efuse_straps_4;

		u32 efuse_straps_3;

		u8 efuse_box_bit_131_124;



		WREG32(RCU_IND_INDEX, 0x204);

		efuse_straps_4 = RREG32(RCU_IND_DATA);

		WREG32(RCU_IND_INDEX, 0x203);

		efuse_straps_3 = RREG32(RCU_IND_DATA);

		efuse_box_bit_131_124 = (u8)(((efuse_straps_4 & 0xf) << 4) | ((efuse_straps_3 & 0xf0000000) >> 28));



		switch(efuse_box_bit_131_124) {

		case 0x00:

			gb_backend_map = 0x76543210;

			break;

		case 0x55:

			gb_backend_map = 0x77553311;

			break;

		case 0x56:

			gb_backend_map = 0x77553300;

			break;

		case 0x59:

			gb_backend_map = 0x77552211;

			break;

		case 0x66:

			gb_backend_map = 0x77443300;

			break;

		case 0x99:

			gb_backend_map = 0x66552211;

			break;

		case 0x5a:

			gb_backend_map = 0x77552200;

			break;

		case 0xaa:

			gb_backend_map = 0x66442200;

			break;

		case 0x95:

			gb_backend_map = 0x66553311;

			break;

		default:

			DRM_ERROR("bad backend map, using default\n");

			gb_backend_map =

				evergreen_get_tile_pipe_to_backend_map(rdev,

								       rdev->config.evergreen.max_tile_pipes,

								       rdev->config.evergreen.max_backends,

								       ((EVERGREEN_MAX_BACKENDS_MASK <<

								   rdev->config.evergreen.max_backends) &

									EVERGREEN_MAX_BACKENDS_MASK));

			break;

		}

	} else if (rdev->ddev->pdev->device == 0x68b9) {

		u32 efuse_straps_3;

		u8 efuse_box_bit_127_124;



		WREG32(RCU_IND_INDEX, 0x203);

		efuse_straps_3 = RREG32(RCU_IND_DATA);

		efuse_box_bit_127_124 = (u8)((efuse_straps_3 & 0xF0000000) >> 28);



		switch(efuse_box_bit_127_124) {

		case 0x0:

			gb_backend_map = 0x00003210;

			break;

		case 0x5:

		case 0x6:

		case 0x9:

		case 0xa:

			gb_backend_map = 0x00003311;

			break;

		default:

			DRM_ERROR("bad backend map, using default\n");

			gb_backend_map =

				evergreen_get_tile_pipe_to_backend_map(rdev,

								       rdev->config.evergreen.max_tile_pipes,

								       rdev->config.evergreen.max_backends,

								       ((EVERGREEN_MAX_BACKENDS_MASK <<

								   rdev->config.evergreen.max_backends) &

									EVERGREEN_MAX_BACKENDS_MASK));

			break;

		}

	} else {

		switch (rdev->family) {

		case CHIP_CYPRESS:

		case CHIP_HEMLOCK:

		case CHIP_BARTS:

			gb_backend_map = 0x66442200;

			break;

		case CHIP_JUNIPER:

			gb_backend_map = 0x00002200;

			break;

		default:

			gb_backend_map =

				evergreen_get_tile_pipe_to_backend_map(rdev,

								       rdev->config.evergreen.max_tile_pipes,

								       rdev->config.evergreen.max_backends,

								       ((EVERGREEN_MAX_BACKENDS_MASK <<

									 rdev->config.evergreen.max_backends) &

									EVERGREEN_MAX_BACKENDS_MASK));

		}

	}



	/* setup tiling info dword.  gb_addr_config is not adequate since it does

	 * not have bank info, so create a custom tiling dword.

	 * bits 3:0   num_pipes
@@ -2147,42 +1863,48 @@ static void evergreen_gpu_init(struct radeon_device *rdev) 
		else

			rdev->config.evergreen.tile_config |= 0 << 4;

	}

	rdev->config.evergreen.tile_config |=

		((mc_arb_ramcfg & BURSTLENGTH_MASK) >> BURSTLENGTH_SHIFT) << 8;

	rdev->config.evergreen.tile_config |= 0 << 8;

	rdev->config.evergreen.tile_config |=

		((gb_addr_config & 0x30000000) >> 28) << 12;



	rdev->config.evergreen.backend_map = gb_backend_map;

	WREG32(GB_BACKEND_MAP, gb_backend_map);

	WREG32(GB_ADDR_CONFIG, gb_addr_config);

	WREG32(DMIF_ADDR_CONFIG, gb_addr_config);

	WREG32(HDP_ADDR_CONFIG, gb_addr_config);

	num_shader_engines = (gb_addr_config & NUM_SHADER_ENGINES(3) >> 12) + 1;



	num_shader_engines = ((RREG32(GB_ADDR_CONFIG) & NUM_SHADER_ENGINES(3)) >> 12) + 1;

	grbm_gfx_index = INSTANCE_BROADCAST_WRITES;

	if ((rdev->family >= CHIP_CEDAR) && (rdev->family <= CHIP_HEMLOCK)) {

		u32 efuse_straps_4;

		u32 efuse_straps_3;



	for (i = 0; i < rdev->config.evergreen.num_ses; i++) {

		u32 rb = cc_rb_backend_disable | (0xf0 << 16);

		u32 sp = cc_gc_shader_pipe_config;

		u32 gfx = grbm_gfx_index | SE_INDEX(i);

		WREG32(RCU_IND_INDEX, 0x204);

		efuse_straps_4 = RREG32(RCU_IND_DATA);

		WREG32(RCU_IND_INDEX, 0x203);

		efuse_straps_3 = RREG32(RCU_IND_DATA);

		tmp = (((efuse_straps_4 & 0xf) << 4) |

		      ((efuse_straps_3 & 0xf0000000) >> 28));

	} else {

		tmp = 0;

		for (i = (rdev->config.evergreen.num_ses - 1); i >= 0; i--) {

			u32 rb_disable_bitmap;



		if (i == num_shader_engines) {

			rb |= BACKEND_DISABLE(EVERGREEN_MAX_BACKENDS_MASK);

			sp |= INACTIVE_SIMDS(EVERGREEN_MAX_SIMDS_MASK);

			WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));

			WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));

			rb_disable_bitmap = (RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000) >> 16;

			tmp <<= 4;

			tmp |= rb_disable_bitmap;

		}

	}

	/* enabled rb are just the one not disabled :) */

	disabled_rb_mask = tmp;



		WREG32(GRBM_GFX_INDEX, gfx);

		WREG32(RLC_GFX_INDEX, gfx);

	WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES);

	WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES);



		WREG32(CC_RB_BACKEND_DISABLE, rb);

		WREG32(CC_SYS_RB_BACKEND_DISABLE, rb);

		WREG32(GC_USER_RB_BACKEND_DISABLE, rb);

		WREG32(CC_GC_SHADER_PIPE_CONFIG, sp);

	}

	WREG32(GB_ADDR_CONFIG, gb_addr_config);

	WREG32(DMIF_ADDR_CONFIG, gb_addr_config);

	WREG32(HDP_ADDR_CONFIG, gb_addr_config);



	grbm_gfx_index = INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES;

	WREG32(GRBM_GFX_INDEX, grbm_gfx_index);

	WREG32(RLC_GFX_INDEX, grbm_gfx_index);

	tmp = gb_addr_config & NUM_PIPES_MASK;

	tmp = r6xx_remap_render_backend(rdev, tmp, rdev->config.evergreen.max_backends,

					EVERGREEN_MAX_BACKENDS, disabled_rb_mask);

	WREG32(GB_BACKEND_MAP, tmp);



	WREG32(CGTS_SYS_TCC_DISABLE, 0);

	WREG32(CGTS_TCC_DISABLE, 0);

drivers/gpu/drm/radeon/evergreend.h

+10 −0
Original line number Diff line number Diff line
@@ -37,6 +37,15 @@ 
#define EVERGREEN_MAX_PIPES_MASK        0xFF

#define EVERGREEN_MAX_LDS_NUM           0xFFFF



#define CYPRESS_GB_ADDR_CONFIG_GOLDEN        0x02011003

#define BARTS_GB_ADDR_CONFIG_GOLDEN          0x02011003

#define CAYMAN_GB_ADDR_CONFIG_GOLDEN         0x02011003

#define JUNIPER_GB_ADDR_CONFIG_GOLDEN        0x02010002

#define REDWOOD_GB_ADDR_CONFIG_GOLDEN        0x02010002

#define TURKS_GB_ADDR_CONFIG_GOLDEN          0x02010002

#define CEDAR_GB_ADDR_CONFIG_GOLDEN          0x02010001

#define CAICOS_GB_ADDR_CONFIG_GOLDEN         0x02010001



/* Registers */



#define RCU_IND_INDEX           			0x100
@@ -54,6 +63,7 @@ 
#define		BACKEND_DISABLE(x)     			((x) << 16)

#define GB_ADDR_CONFIG  				0x98F8

#define		NUM_PIPES(x)				((x) << 0)

#define		NUM_PIPES_MASK				0x0000000f

#define		PIPE_INTERLEAVE_SIZE(x)			((x) << 4)

#define		BANK_INTERLEAVE_SIZE(x)			((x) << 8)

#define		NUM_SHADER_ENGINES(x)			((x) << 12)

drivers/gpu/drm/radeon/ni.c

+29 −322

File changed.

Preview size limit exceeded, changes collapsed.

drivers/gpu/drm/radeon/nid.h

+11 −0
Original line number Diff line number Diff line
@@ -41,6 +41,9 @@ 
#define CAYMAN_MAX_TCC               16

#define CAYMAN_MAX_TCC_MASK          0xFF



#define CAYMAN_GB_ADDR_CONFIG_GOLDEN       0x02011003

#define ARUBA_GB_ADDR_CONFIG_GOLDEN        0x12010001



#define DMIF_ADDR_CONFIG  				0xBD4

#define	SRBM_GFX_CNTL				        0x0E44

#define		RINGID(x)					(((x) & 0x3) << 0)
@@ -148,6 +151,8 @@ 
#define	CGTS_SYS_TCC_DISABLE				0x3F90

#define	CGTS_USER_SYS_TCC_DISABLE			0x3F94



#define RLC_GFX_INDEX           			0x3FC4



#define	CONFIG_MEMSIZE					0x5428



#define HDP_MEM_COHERENCY_FLUSH_CNTL			0x5480
@@ -212,6 +217,12 @@ 
#define		SOFT_RESET_VGT					(1 << 14)

#define		SOFT_RESET_IA					(1 << 15)



#define GRBM_GFX_INDEX          			0x802C

#define		INSTANCE_INDEX(x)			((x) << 0)

#define		SE_INDEX(x)     			((x) << 16)

#define		INSTANCE_BROADCAST_WRITES      		(1 << 30)

#define		SE_BROADCAST_WRITES      		(1 << 31)



#define	SCRATCH_REG0					0x8500

#define	SCRATCH_REG1					0x8504

#define	SCRATCH_REG2					0x8508

drivers/gpu/drm/radeon/r600.c

+69 −130
Original line number Diff line number Diff line
@@ -1376,113 +1376,51 @@ int r600_asic_reset(struct radeon_device *rdev) 
	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 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;



	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;



	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;

		}

		if (enabled_backends_count == num_backends)

			break;

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 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;



	/* mask out the RBs that don't exist on that asic */

	disabled_rb_mask |= (0xff << max_rb_num) & 0xff;



	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;

	}



	if (enabled_backends_count == 0) {

		enabled_backends_mask = 1;

		enabled_backends_count = 1;

	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 (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;

			if (pipe_rb_remain) {

				data <<= rb_num_width;

				data |= max_rb_num - i - 1;

				pipe_rb_remain--;

			}



	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;

		}

		mask >>= 1;

	}



	return backend_map;

	return data;

}



int r600_count_pipe_bits(uint32_t val)
@@ -1500,7 +1438,6 @@ void r600_gpu_init(struct radeon_device *rdev) 
{

	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 @@ void r600_gpu_init(struct radeon_device *rdev) 
	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 @@ void r600_gpu_init(struct radeon_device *rdev) 
	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 @@ void r600_gpu_init(struct radeon_device *rdev) 
	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);



	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);



	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));

	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) & 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;

	}



	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);

CRA Git | Maintained and supported by SUSTech CRA and CCSE