Subversion Repositories Kolibri OS

/* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */

/*

 * Copyright (C) 2013 Rob Clark 

 *

 * Permission is hereby granted, free of charge, to any person obtaining a

 * copy of this software and associated documentation files (the "Software"),

 * to deal in the Software without restriction, including without limitation

 * the rights to use, copy, modify, merge, publish, distribute, sublicense,

 * and/or sell copies of the Software, and to permit persons to whom the

 * Software is furnished to do so, subject to the following conditions:

 *

 * The above copyright notice and this permission notice (including the next

 * paragraph) shall be included in all copies or substantial portions of the

 * Software.

 *

 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL

 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER

 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,

 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE

 * SOFTWARE.

 *

 * Authors:

 *    Rob Clark 

 */

#include "pipe/p_state.h"

#include "util/u_string.h"

#include "util/u_memory.h"

#include "util/u_helpers.h"

#include "util/u_format.h"

#include "freedreno_resource.h"

#include "fd3_emit.h"

#include "fd3_blend.h"

#include "fd3_context.h"

#include "fd3_program.h"

#include "fd3_rasterizer.h"

#include "fd3_texture.h"

#include "fd3_util.h"

#include "fd3_zsa.h"

/* regid:          base const register

 * prsc or dwords: buffer containing constant values

 * sizedwords:     size of const value buffer

 */

void

fd3_emit_constant(struct fd_ringbuffer *ring,

		enum adreno_state_block sb,

		uint32_t regid, uint32_t offset, uint32_t sizedwords,

		const uint32_t *dwords, struct pipe_resource *prsc)

{

	uint32_t i, sz;

	enum adreno_state_src src;

	if (prsc) {

		sz = 0;

		src = SS_INDIRECT;

	} else {

		sz = sizedwords;

		src = SS_DIRECT;

	}

	/* we have this sometimes, not others.. perhaps we could be clever

	 * and figure out actually when we need to invalidate cache:

	 */

	OUT_PKT0(ring, REG_A3XX_UCHE_CACHE_INVALIDATE0_REG, 2);

	OUT_RING(ring, A3XX_UCHE_CACHE_INVALIDATE0_REG_ADDR(0));

	OUT_RING(ring, A3XX_UCHE_CACHE_INVALIDATE1_REG_ADDR(0) |

			A3XX_UCHE_CACHE_INVALIDATE1_REG_OPCODE(INVALIDATE) |

			A3XX_UCHE_CACHE_INVALIDATE1_REG_ENTIRE_CACHE);

	OUT_PKT3(ring, CP_LOAD_STATE, 2 + sz);

	OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/2) |

			CP_LOAD_STATE_0_STATE_SRC(src) |

			CP_LOAD_STATE_0_STATE_BLOCK(sb) |

			CP_LOAD_STATE_0_NUM_UNIT(sizedwords/2));

	if (prsc) {

		struct fd_bo *bo = fd_resource(prsc)->bo;

		OUT_RELOC(ring, bo, offset,

				CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS), 0);

	} else {

		OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) |

				CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS));

		dwords = (uint32_t *)&((uint8_t *)dwords)[offset];

	}

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

		OUT_RING(ring, dwords[i]);

	}

}

static void

emit_constants(struct fd_ringbuffer *ring,

		enum adreno_state_block sb,

		struct fd_constbuf_stateobj *constbuf,

		struct fd3_shader_stateobj *shader)

{

	uint32_t enabled_mask = constbuf->enabled_mask;

	uint32_t base = 0;

	unsigned i;

	// XXX TODO only emit dirty consts.. but we need to keep track if

	// they are clobbered by a clear, gmem2mem, or mem2gmem..

	constbuf->dirty_mask = enabled_mask;

	/* emit user constants: */

	while (enabled_mask) {

		unsigned index = ffs(enabled_mask) - 1;

		struct pipe_constant_buffer *cb = &constbuf->cb[index];

		unsigned size = align(cb->buffer_size, 4) / 4; /* size in dwords */

		// I expect that size should be a multiple of vec4's:

		assert(size == align(size, 4));

		/* gallium could have const-buffer still bound, even though the

		 * shader is not using it.  Writing consts above constlen (or

		 * rather, HLSQ_{VS,FS}_CONTROL_REG.CONSTLENGTH) will cause a

		 * hang.

		 */

		if ((base / 4) >= shader->constlen)

			break;

		if (constbuf->dirty_mask & (1 << index)) {

			fd3_emit_constant(ring, sb, base,

					cb->buffer_offset, size,

					cb->user_buffer, cb->buffer);

			constbuf->dirty_mask &= ~(1 << index);

		}

		base += size;

		enabled_mask &= ~(1 << index);

	}

	/* emit shader immediates: */

	if (shader) {

		for (i = 0; i < shader->immediates_count; i++) {

			fd3_emit_constant(ring, sb,

					4 * (shader->first_immediate + i),

					0, 4, shader->immediates[i].val, NULL);

		}

	}

}

#define VERT_TEX_OFF    0

#define FRAG_TEX_OFF    16

#define BASETABLE_SZ    14

static void

emit_textures(struct fd_ringbuffer *ring,

		enum adreno_state_block sb,

		struct fd_texture_stateobj *tex)

{

	static const unsigned tex_off[] = {

			[SB_VERT_TEX] = VERT_TEX_OFF,

			[SB_FRAG_TEX] = FRAG_TEX_OFF,

	};

	static const enum adreno_state_block mipaddr[] = {

			[SB_VERT_TEX] = SB_VERT_MIPADDR,

			[SB_FRAG_TEX] = SB_FRAG_MIPADDR,

	};

	unsigned i, j;

	assert(tex->num_samplers == tex->num_textures);  // TODO check..

	if (!tex->num_samplers)

		return;

	/* output sampler state: */

	OUT_PKT3(ring, CP_LOAD_STATE, 2 + (2 * tex->num_samplers));

	OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(tex_off[sb]) |

			CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |

			CP_LOAD_STATE_0_STATE_BLOCK(sb) |

			CP_LOAD_STATE_0_NUM_UNIT(tex->num_samplers));

	OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) |

			CP_LOAD_STATE_1_EXT_SRC_ADDR(0));

	for (i = 0; i < tex->num_samplers; i++) {

		struct fd3_sampler_stateobj *sampler =

				fd3_sampler_stateobj(tex->samplers[i]);

		OUT_RING(ring, sampler->texsamp0);

		OUT_RING(ring, sampler->texsamp1);

	}

	/* emit texture state: */

	OUT_PKT3(ring, CP_LOAD_STATE, 2 + (4 * tex->num_textures));

	OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(tex_off[sb]) |

			CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |

			CP_LOAD_STATE_0_STATE_BLOCK(sb) |

			CP_LOAD_STATE_0_NUM_UNIT(tex->num_textures));

	OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |

			CP_LOAD_STATE_1_EXT_SRC_ADDR(0));

	for (i = 0; i < tex->num_textures; i++) {

		struct fd3_pipe_sampler_view *view =

				fd3_pipe_sampler_view(tex->textures[i]);

		OUT_RING(ring, view->texconst0);

		OUT_RING(ring, view->texconst1);

		OUT_RING(ring, view->texconst2 |

				A3XX_TEX_CONST_2_INDX(BASETABLE_SZ * i));

		OUT_RING(ring, view->texconst3);

	}

	/* emit mipaddrs: */

	OUT_PKT3(ring, CP_LOAD_STATE, 2 + (BASETABLE_SZ * tex->num_textures));

	OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(BASETABLE_SZ * tex_off[sb]) |

			CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) |

			CP_LOAD_STATE_0_STATE_BLOCK(mipaddr[sb]) |

			CP_LOAD_STATE_0_NUM_UNIT(BASETABLE_SZ * tex->num_textures));

	OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) |

			CP_LOAD_STATE_1_EXT_SRC_ADDR(0));

	for (i = 0; i < tex->num_textures; i++) {

		struct fd3_pipe_sampler_view *view =

				fd3_pipe_sampler_view(tex->textures[i]);

		OUT_RELOC(ring, view->tex_resource->bo, 0, 0, 0);

		/* I think each entry is a ptr to mipmap level.. for now, just

		 * pad w/ null's until I get around to actually implementing

		 * mipmap support..

		 */

		for (j = 1; j < BASETABLE_SZ; j++) {

			OUT_RING(ring, 0x00000000);

		}

	}

}

static void

emit_cache_flush(struct fd_ringbuffer *ring)

{

	OUT_PKT3(ring, CP_EVENT_WRITE, 1);

	OUT_RING(ring, CACHE_FLUSH);

	OUT_PKT3(ring, CP_DRAW_INDX, 3);

	OUT_RING(ring, 0x00000000);

	OUT_RING(ring, DRAW(DI_PT_POINTLIST, DI_SRC_SEL_AUTO_INDEX,

			INDEX_SIZE_IGN, IGNORE_VISIBILITY));

	OUT_RING(ring, 0);					/* NumIndices */

	OUT_PKT3(ring, CP_NOP, 4);

	OUT_RING(ring, 0x00000000);

	OUT_RING(ring, 0x00000000);

	OUT_RING(ring, 0x00000000);

	OUT_RING(ring, 0x00000000);

	OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1);

	OUT_RING(ring, 0x00000000);

}

/* emit texture state for mem->gmem restore operation.. eventually it would

 * be good to get rid of this and use normal CSO/etc state for more of these

 * special cases, but for now the compiler is not sufficient..

 */

void

fd3_emit_gmem_restore_tex(struct fd_ringbuffer *ring, struct pipe_surface *psurf)

{

	struct fd_resource *rsc = fd_resource(psurf->texture);