/* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
/*
- * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
+ * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* Rob Clark <robclark@freedesktop.org>
*/
-#include "pipe/p_state.h"
-#include "util/u_string.h"
-#include "util/u_memory.h"
-#include "util/u_inlines.h"
-#include "util/u_format.h"
-#include "tgsi/tgsi_dump.h"
-#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_text.h"
+#include "tgsi/tgsi_ureg.h"
#include "freedreno_program.h"
-#include "freedreno_compiler.h"
-#include "freedreno_vbo.h"
-#include "freedreno_texture.h"
-#include "freedreno_util.h"
-
-static struct fd_shader_stateobj *
-create_shader(enum shader_t type)
-{
- struct fd_shader_stateobj *so = CALLOC_STRUCT(fd_shader_stateobj);
- if (!so)
- return NULL;
- so->type = type;
- return so;
-}
-
-static void
-delete_shader(struct fd_shader_stateobj *so)
-{
- ir_shader_destroy(so->ir);
- FREE(so->tokens);
- FREE(so);
-}
-
-static struct fd_shader_stateobj *
-assemble(struct fd_shader_stateobj *so)
-{
- free(so->bin);
- so->bin = ir_shader_assemble(so->ir, &so->info);
- if (!so->bin)
- goto fail;
-
- if (fd_mesa_debug & FD_DBG_DISASM) {
- DBG("disassemble: type=%d", so->type);
- disasm(so->bin, so->info.sizedwords, 0, so->type);
- }
-
- return so;
-
-fail:
- debug_error("assemble failed!");
- delete_shader(so);
- return NULL;
-}
-
-static struct fd_shader_stateobj *
-compile(struct fd_program_stateobj *prog, struct fd_shader_stateobj *so)
-{
- int ret;
-
- if (fd_mesa_debug & FD_DBG_DISASM) {
- DBG("dump tgsi: type=%d", so->type);
- tgsi_dump(so->tokens, 0);
- }
-
- ret = fd_compile_shader(prog, so);
- if (ret)
- goto fail;
-
- /* NOTE: we don't assemble yet because for VS we don't know the
- * type information for vertex fetch yet.. so those need to be
- * patched up later before assembling.
- */
-
- so->info.sizedwords = 0;
-
- return so;
-
-fail:
- debug_error("compile failed!");
- delete_shader(so);
- return NULL;
-}
-
-static void
-emit(struct fd_ringbuffer *ring, struct fd_shader_stateobj *so)
-{
- unsigned i;
-
- if (so->info.sizedwords == 0)
- assemble(so);
-
- OUT_PKT3(ring, CP_IM_LOAD_IMMEDIATE, 2 + so->info.sizedwords);
- OUT_RING(ring, (so->type == SHADER_VERTEX) ? 0 : 1);
- OUT_RING(ring, so->info.sizedwords);
- for (i = 0; i < so->info.sizedwords; i++)
- OUT_RING(ring, so->bin[i]);
-}
-
-static void *
-fd_fp_state_create(struct pipe_context *pctx,
- const struct pipe_shader_state *cso)
-{
- struct fd_shader_stateobj *so = create_shader(SHADER_FRAGMENT);
- if (!so)
- return NULL;
- so->tokens = tgsi_dup_tokens(cso->tokens);
- return so;
-}
-
-static void
-fd_fp_state_delete(struct pipe_context *pctx, void *hwcso)
-{
- struct fd_shader_stateobj *so = hwcso;
- delete_shader(so);
-}
+#include "freedreno_context.h"
static void
fd_fp_state_bind(struct pipe_context *pctx, void *hwcso)
ctx->dirty |= FD_DIRTY_PROG;
}
-static void *
-fd_vp_state_create(struct pipe_context *pctx,
- const struct pipe_shader_state *cso)
-{
- struct fd_shader_stateobj *so = create_shader(SHADER_VERTEX);
- if (!so)
- return NULL;
- so->tokens = tgsi_dup_tokens(cso->tokens);
- return so;
-}
-
-static void
-fd_vp_state_delete(struct pipe_context *pctx, void *hwcso)
-{
- struct fd_shader_stateobj *so = hwcso;
- delete_shader(so);
-}
-
static void
fd_vp_state_bind(struct pipe_context *pctx, void *hwcso)
{
ctx->dirty |= FD_DIRTY_PROG;
}
-static void
-patch_vtx_fetches(struct fd_context *ctx, struct fd_shader_stateobj *so,
- struct fd_vertex_stateobj *vtx)
-{
- unsigned i;
-
- assert(so->num_vfetch_instrs == vtx->num_elements);
-
- /* update vtx fetch instructions: */
- for (i = 0; i < so->num_vfetch_instrs; i++) {
- struct ir_instruction *instr = so->vfetch_instrs[i];
- struct pipe_vertex_element *elem = &vtx->pipe[i];
- struct pipe_vertex_buffer *vb =
- &ctx->vertexbuf.vb[elem->vertex_buffer_index];
- enum pipe_format format = elem->src_format;
- const struct util_format_description *desc =
- util_format_description(format);
- unsigned j;
-
- /* Find the first non-VOID channel. */
- for (j = 0; j < 4; j++)
- if (desc->channel[j].type != UTIL_FORMAT_TYPE_VOID)
- break;
-
- /* CI/CIS can probably be set in compiler instead: */
- instr->fetch.const_idx = 20 + (i / 3);
- instr->fetch.const_idx_sel = i % 3;
-
- instr->fetch.fmt = fd_pipe2surface(format);
- instr->fetch.is_normalized = desc->channel[j].normalized;
- instr->fetch.is_signed =
- desc->channel[j].type == UTIL_FORMAT_TYPE_SIGNED;
- instr->fetch.stride = vb->stride ? : 1;
- instr->fetch.offset = elem->src_offset;
+static const char *solid_fp =
+ "FRAG \n"
+ "PROPERTY FS_COLOR0_WRITES_ALL_CBUFS 1 \n"
+ "DCL CONST[0] \n"
+ "DCL OUT[0], COLOR \n"
+ " 0: MOV OUT[0], CONST[0] \n"
+ " 1: END \n";
- for (j = 0; j < 4; j++)
- instr->regs[0]->swizzle[j] = "xyzw01__"[desc->swizzle[j]];
+static const char *solid_vp =
+ "VERT \n"
+ "DCL IN[0] \n"
+ "DCL OUT[0], POSITION \n"
+ " 0: MOV OUT[0], IN[0] \n"
+ " 1: END \n";
- assert(instr->fetch.fmt != FMT_INVALID);
+static const char *blit_vp =
+ "VERT \n"
+ "DCL IN[0] \n"
+ "DCL IN[1] \n"
+ "DCL OUT[0], TEXCOORD[0] \n"
+ "DCL OUT[1], POSITION \n"
+ " 0: MOV OUT[0], IN[0] \n"
+ " 0: MOV OUT[1], IN[1] \n"
+ " 1: END \n";
- DBG("vtx[%d]: %s (%d), ci=%d, cis=%d, id=%d, swizzle=%s, "
- "stride=%d, offset=%d",
- i, util_format_name(format),
- instr->fetch.fmt,
- instr->fetch.const_idx,
- instr->fetch.const_idx_sel,
- elem->instance_divisor,
- instr->regs[0]->swizzle,
- instr->fetch.stride,
- instr->fetch.offset);
- }
-
- /* trigger re-assemble: */
- so->info.sizedwords = 0;
-}
-
-static void
-patch_tex_fetches(struct fd_context *ctx, struct fd_shader_stateobj *so,
- struct fd_texture_stateobj *tex)
+static void * assemble_tgsi(struct pipe_context *pctx,
+ const char *src, bool frag)
{
- unsigned i;
+ struct tgsi_token toks[32];
+ struct pipe_shader_state cso = {
+ .tokens = toks,
+ };
- /* update tex fetch instructions: */
- for (i = 0; i < so->num_tfetch_instrs; i++) {
- struct ir_instruction *instr = so->tfetch_instrs[i].instr;
- unsigned samp_id = so->tfetch_instrs[i].samp_id;
- unsigned const_idx = fd_get_const_idx(ctx, tex, samp_id);
+ tgsi_text_translate(src, toks, ARRAY_SIZE(toks));
- if (const_idx != instr->fetch.const_idx) {
- instr->fetch.const_idx = const_idx;
- /* trigger re-assemble: */
- so->info.sizedwords = 0;
- }
- }
+ if (frag)
+ return pctx->create_fs_state(pctx, &cso);
+ else
+ return pctx->create_vs_state(pctx, &cso);
}
-void
-fd_program_validate(struct fd_context *ctx)
-{
- struct fd_program_stateobj *prog = &ctx->prog;
-
- /* if vertex or frag shader is dirty, we may need to recompile. Compile
- * frag shader first, as that assigns the register slots for exports
- * from the vertex shader. And therefore if frag shader has changed we
- * need to recompile both vert and frag shader.
- */
- if (prog->dirty & FD_SHADER_DIRTY_FP)
- compile(prog, prog->fp);
-
- if (prog->dirty & (FD_SHADER_DIRTY_FP | FD_SHADER_DIRTY_VP))
- compile(prog, prog->vp);
-
- if (prog->dirty)
- ctx->dirty |= FD_DIRTY_PROG;
-
- prog->dirty = 0;
-
- /* if necessary, fix up vertex fetch instructions: */
- if (ctx->dirty & (FD_DIRTY_VTX | FD_DIRTY_VERTEXBUF | FD_DIRTY_PROG))
- patch_vtx_fetches(ctx, prog->vp, ctx->vtx);
-
- /* if necessary, fix up texture fetch instructions: */
- if (ctx->dirty & (FD_DIRTY_VERTTEX | FD_DIRTY_FRAGTEX | FD_DIRTY_PROG)) {
- patch_tex_fetches(ctx, prog->vp, &ctx->verttex);
- patch_tex_fetches(ctx, prog->fp, &ctx->fragtex);
- }
-}
-
-void
-fd_program_emit(struct fd_ringbuffer *ring,
- struct fd_program_stateobj *prog)
-{
- struct ir_shader_info *vsi = &prog->vp->info;
- struct ir_shader_info *fsi = &prog->fp->info;
- uint8_t vs_gprs, fs_gprs, vs_export;
-
- emit(ring, prog->vp);
- emit(ring, prog->fp);
-
- vs_gprs = (vsi->max_reg < 0) ? 0x80 : vsi->max_reg;
- fs_gprs = (fsi->max_reg < 0) ? 0x80 : fsi->max_reg;
- vs_export = max(1, prog->num_exports) - 1;
-
- OUT_PKT3(ring, CP_SET_CONSTANT, 2);
- OUT_RING(ring, CP_REG(REG_SQ_PROGRAM_CNTL));
- OUT_RING(ring, SQ_PROGRAM_CNTL_PS_EXPORT_MODE(POSITION_2_VECTORS_SPRITE) |
- SQ_PROGRAM_CNTL_VS_RESOURCE |
- SQ_PROGRAM_CNTL_PS_RESOURCE |
- SQ_PROGRAM_CNTL_VS_EXPORT_COUNT(vs_export) |
- SQ_PROGRAM_CNTL_PS_REGS(fs_gprs) |
- SQ_PROGRAM_CNTL_VS_REGS(vs_gprs));
-}
-
-/* Creates shader:
- * EXEC ADDR(0x2) CNT(0x1)
- * (S)FETCH: SAMPLE R0.xyzw = R0.xyx CONST(0) LOCATION(CENTER)
- * ALLOC PARAM/PIXEL SIZE(0x0)
- * EXEC_END ADDR(0x3) CNT(0x1)
- * ALU: MAXv export0 = R0, R0 ; gl_FragColor
- * NOP
- */
-static struct fd_shader_stateobj *
-create_blit_fp(void)
-{
- struct fd_shader_stateobj *so = create_shader(SHADER_FRAGMENT);
- struct ir_cf *cf;
- struct ir_instruction *instr;
-
- if (!so)
- return NULL;
-
- so->ir = ir_shader_create();
-
- cf = ir_cf_create(so->ir, EXEC);
-
- instr = ir_instr_create_tex_fetch(cf, 0);
- ir_reg_create(instr, 0, "xyzw", 0);
- ir_reg_create(instr, 0, "xyx", 0);
- instr->sync = true;
-
- cf = ir_cf_create_alloc(so->ir, SQ_PARAMETER_PIXEL, 0);
- cf = ir_cf_create(so->ir, EXEC_END);
-
- instr = ir_instr_create_alu(cf, MAXv, ~0);
- ir_reg_create(instr, 0, NULL, IR_REG_EXPORT);
- ir_reg_create(instr, 0, NULL, 0);
- ir_reg_create(instr, 0, NULL, 0);
-
- return assemble(so);
-}
-
-/* Creates shader:
-* EXEC ADDR(0x3) CNT(0x2)
-* FETCH: VERTEX R1.xy01 = R0.x FMT_32_32_FLOAT UNSIGNED STRIDE(8) CONST(26, 1)
-* FETCH: VERTEX R2.xyz1 = R0.x FMT_32_32_32_FLOAT UNSIGNED STRIDE(12) CONST(26, 0)
-* ALLOC POSITION SIZE(0x0)
-* EXEC ADDR(0x5) CNT(0x1)
-* ALU: MAXv export62 = R2, R2 ; gl_Position
-* ALLOC PARAM/PIXEL SIZE(0x0)
-* EXEC_END ADDR(0x6) CNT(0x1)
-* ALU: MAXv export0 = R1, R1
-* NOP
- */
-static struct fd_shader_stateobj *
-create_blit_vp(void)
-{
- struct fd_shader_stateobj *so = create_shader(SHADER_VERTEX);
- struct ir_cf *cf;
- struct ir_instruction *instr;
-
- if (!so)
- return NULL;
-
- so->ir = ir_shader_create();
-
- cf = ir_cf_create(so->ir, EXEC);
-
- instr = ir_instr_create_vtx_fetch(cf, 26, 1, FMT_32_32_FLOAT, false, 8);
- instr->fetch.is_normalized = true;
- ir_reg_create(instr, 1, "xy01", 0);
- ir_reg_create(instr, 0, "x", 0);
-
- instr = ir_instr_create_vtx_fetch(cf, 26, 0, FMT_32_32_32_FLOAT, false, 12);
- instr->fetch.is_normalized = true;
- ir_reg_create(instr, 2, "xyz1", 0);
- ir_reg_create(instr, 0, "x", 0);
-
- cf = ir_cf_create_alloc(so->ir, SQ_POSITION, 0);
- cf = ir_cf_create(so->ir, EXEC);
-
- instr = ir_instr_create_alu(cf, MAXv, ~0);
- ir_reg_create(instr, 62, NULL, IR_REG_EXPORT);
- ir_reg_create(instr, 2, NULL, 0);
- ir_reg_create(instr, 2, NULL, 0);
-
- cf = ir_cf_create_alloc(so->ir, SQ_PARAMETER_PIXEL, 0);
- cf = ir_cf_create(so->ir, EXEC_END);
-
- instr = ir_instr_create_alu(cf, MAXv, ~0);
- ir_reg_create(instr, 0, NULL, IR_REG_EXPORT);
- ir_reg_create(instr, 1, NULL, 0);
- ir_reg_create(instr, 1, NULL, 0);
-
-
- return assemble(so);
-
-}
-
-/* Creates shader:
- * ALLOC PARAM/PIXEL SIZE(0x0)
- * EXEC_END ADDR(0x1) CNT(0x1)
- * ALU: MAXv export0 = C0, C0 ; gl_FragColor
- */
-static struct fd_shader_stateobj *
-create_solid_fp(void)
+static void *
+fd_prog_blit(struct pipe_context *pctx, int rts, bool depth)
{
- struct fd_shader_stateobj *so = create_shader(SHADER_FRAGMENT);
- struct ir_cf *cf;
- struct ir_instruction *instr;
-
- if (!so)
+ int i;
+ struct ureg_src tc;
+ struct ureg_program *ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
+ if (!ureg)
return NULL;
- so->ir = ir_shader_create();
+ tc = ureg_DECL_fs_input(
+ ureg, TGSI_SEMANTIC_GENERIC, 0, TGSI_INTERPOLATE_PERSPECTIVE);
+ for (i = 0; i < rts; i++)
+ ureg_TEX(ureg, ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, i),
+ TGSI_TEXTURE_2D, tc, ureg_DECL_sampler(ureg, i));
+ if (depth)
+ ureg_TEX(ureg,
+ ureg_writemask(
+ ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0),
+ TGSI_WRITEMASK_Z),
+ TGSI_TEXTURE_2D, tc, ureg_DECL_sampler(ureg, rts));
- cf = ir_cf_create_alloc(so->ir, SQ_PARAMETER_PIXEL, 0);
- cf = ir_cf_create(so->ir, EXEC_END);
+ ureg_END(ureg);
- instr = ir_instr_create_alu(cf, MAXv, ~0);
- ir_reg_create(instr, 0, NULL, IR_REG_EXPORT);
- ir_reg_create(instr, 0, NULL, IR_REG_CONST);
- ir_reg_create(instr, 0, NULL, IR_REG_CONST);
-
- return assemble(so);
+ return ureg_create_shader_and_destroy(ureg, pctx);
}
-/* Creates shader:
- * EXEC ADDR(0x3) CNT(0x1)
- * (S)FETCH: VERTEX R1.xyz1 = R0.x FMT_32_32_32_FLOAT
- * UNSIGNED STRIDE(12) CONST(26, 0)
- * ALLOC POSITION SIZE(0x0)
- * EXEC ADDR(0x4) CNT(0x1)
- * ALU: MAXv export62 = R1, R1 ; gl_Position
- * ALLOC PARAM/PIXEL SIZE(0x0)
- * EXEC_END ADDR(0x5) CNT(0x0)
- */
-static struct fd_shader_stateobj *
-create_solid_vp(void)
-{
- struct fd_shader_stateobj *so = create_shader(SHADER_VERTEX);
- struct ir_cf *cf;
- struct ir_instruction *instr;
-
- if (!so)
- return NULL;
-
- so->ir = ir_shader_create();
-
- cf = ir_cf_create(so->ir, EXEC);
-
- instr = ir_instr_create_vtx_fetch(cf, 26, 0, FMT_32_32_32_FLOAT, false, 12);
- ir_reg_create(instr, 1, "xyz1", 0);
- ir_reg_create(instr, 0, "x", 0);
-
- cf = ir_cf_create_alloc(so->ir, SQ_POSITION, 0);
- cf = ir_cf_create(so->ir, EXEC);
-
- instr = ir_instr_create_alu(cf, MAXv, ~0);
- ir_reg_create(instr, 62, NULL, IR_REG_EXPORT);
- ir_reg_create(instr, 1, NULL, 0);
- ir_reg_create(instr, 1, NULL, 0);
- cf = ir_cf_create_alloc(so->ir, SQ_PARAMETER_PIXEL, 0);
- cf = ir_cf_create(so->ir, EXEC_END);
-
- return assemble(so);
-}
-
-void
-fd_prog_init(struct pipe_context *pctx)
+void fd_prog_init(struct pipe_context *pctx)
{
struct fd_context *ctx = fd_context(pctx);
+ int i;
- pctx->create_fs_state = fd_fp_state_create;
pctx->bind_fs_state = fd_fp_state_bind;
- pctx->delete_fs_state = fd_fp_state_delete;
-
- pctx->create_vs_state = fd_vp_state_create;
pctx->bind_vs_state = fd_vp_state_bind;
- pctx->delete_vs_state = fd_vp_state_delete;
- ctx->solid_prog.fp = create_solid_fp();
- ctx->solid_prog.vp = create_solid_vp();
- ctx->blit_prog.fp = create_blit_fp();
- ctx->blit_prog.vp = create_blit_vp();
+ // XXX for now, let a2xx keep it's own hand-rolled shaders
+ // for solid and blit progs:
+ if (ctx->screen->gpu_id < 300)
+ return;
+
+ ctx->solid_prog.fp = assemble_tgsi(pctx, solid_fp, true);
+ ctx->solid_prog.vp = assemble_tgsi(pctx, solid_vp, false);
+ ctx->blit_prog[0].vp = assemble_tgsi(pctx, blit_vp, false);
+ ctx->blit_prog[0].fp = fd_prog_blit(pctx, 1, false);
+ for (i = 1; i < ctx->screen->max_rts; i++) {
+ ctx->blit_prog[i].vp = ctx->blit_prog[0].vp;
+ ctx->blit_prog[i].fp = fd_prog_blit(pctx, i + 1, false);
+ }
+
+ ctx->blit_z.vp = ctx->blit_prog[0].vp;
+ ctx->blit_z.fp = fd_prog_blit(pctx, 0, true);
+ ctx->blit_zs.vp = ctx->blit_prog[0].vp;
+ ctx->blit_zs.fp = fd_prog_blit(pctx, 1, true);
}
-void
-fd_prog_fini(struct pipe_context *pctx)
+void fd_prog_fini(struct pipe_context *pctx)
{
struct fd_context *ctx = fd_context(pctx);
-
- delete_shader(ctx->solid_prog.vp);
- delete_shader(ctx->solid_prog.fp);
- delete_shader(ctx->blit_prog.vp);
- delete_shader(ctx->blit_prog.fp);
+ int i;
+
+ pctx->delete_vs_state(pctx, ctx->solid_prog.vp);
+ pctx->delete_fs_state(pctx, ctx->solid_prog.fp);
+ pctx->delete_vs_state(pctx, ctx->blit_prog[0].vp);
+ for (i = 0; i < ctx->screen->max_rts; i++)
+ pctx->delete_fs_state(pctx, ctx->blit_prog[i].fp);
+ pctx->delete_fs_state(pctx, ctx->blit_z.fp);
+ pctx->delete_fs_state(pctx, ctx->blit_zs.fp);
}