#include "r600_formats.h"
#include "r600_opcodes.h"
#include "r600_shader.h"
+#include "r600_dump.h"
#include "r600d.h"
+#include "sfn/sfn_nir.h"
#include "sb/sb_public.h"
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_scan.h"
#include "tgsi/tgsi_dump.h"
+#include "tgsi/tgsi_from_mesa.h"
+#include "nir/tgsi_to_nir.h"
+#include "nir/nir_to_tgsi_info.h"
+#include "compiler/nir/nir.h"
#include "util/u_bitcast.h"
#include "util/u_memory.h"
#include "util/u_math.h"
if (shader->bo == NULL) {
return -ENOMEM;
}
- ptr = r600_buffer_map_sync_with_rings(&rctx->b, shader->bo, PIPE_TRANSFER_WRITE);
+ ptr = r600_buffer_map_sync_with_rings(
+ &rctx->b, shader->bo,
+ PIPE_TRANSFER_WRITE | RADEON_TRANSFER_TEMPORARY);
if (R600_BIG_ENDIAN) {
for (i = 0; i < shader->shader.bc.ndw; ++i) {
ptr[i] = util_cpu_to_le32(shader->shader.bc.bytecode[i]);
return 0;
}
+extern const struct nir_shader_compiler_options r600_nir_options;
+static int nshader = 0;
int r600_pipe_shader_create(struct pipe_context *ctx,
struct r600_pipe_shader *shader,
union r600_shader_key key)
struct r600_context *rctx = (struct r600_context *)ctx;
struct r600_pipe_shader_selector *sel = shader->selector;
int r;
- bool dump = r600_can_dump_shader(&rctx->screen->b,
- tgsi_get_processor_type(sel->tokens));
- unsigned use_sb = !(rctx->screen->b.debug_flags & DBG_NO_SB);
+ struct r600_screen *rscreen = (struct r600_screen *)ctx->screen;
+
+ int processor = sel->ir_type == PIPE_SHADER_IR_TGSI ?
+ tgsi_get_processor_type(sel->tokens):
+ pipe_shader_type_from_mesa(sel->nir->info.stage);
+
+ bool dump = r600_can_dump_shader(&rctx->screen->b, processor);
+ unsigned use_sb = !(rctx->screen->b.debug_flags & DBG_NO_SB) &&
+ !(rscreen->b.debug_flags & DBG_NIR);
unsigned sb_disasm;
unsigned export_shader;
-
+
shader->shader.bc.isa = rctx->isa;
+
+ if (!(rscreen->b.debug_flags & DBG_NIR)) {
+ assert(sel->ir_type == PIPE_SHADER_IR_TGSI);
+ r = r600_shader_from_tgsi(rctx, shader, key);
+ if (r) {
+ R600_ERR("translation from TGSI failed !\n");
+ goto error;
+ }
+ } else {
+ if (sel->ir_type == PIPE_SHADER_IR_TGSI) {
+ sel->nir = tgsi_to_nir(sel->tokens, ctx->screen, true);
+ /* Lower int64 ops because we have some r600 build-in shaders that use it */
+ if (!ctx->screen->get_param(ctx->screen, PIPE_CAP_DOUBLES)) {
+ NIR_PASS_V(sel->nir, nir_lower_regs_to_ssa);
+ NIR_PASS_V(sel->nir, nir_lower_alu_to_scalar, NULL, NULL);
+ NIR_PASS_V(sel->nir, nir_lower_int64);
+ NIR_PASS_V(sel->nir, nir_opt_vectorize, NULL, NULL);
+ }
+ NIR_PASS_V(sel->nir, nir_lower_flrp, ~0, false);
+ }
+ nir_tgsi_scan_shader(sel->nir, &sel->info, true);
+ r = r600_shader_from_nir(rctx, shader, &key);
+ if (r) {
+ fprintf(stderr, "--Failed shader--------------------------------------------------\n");
+
+ if (sel->ir_type == PIPE_SHADER_IR_TGSI) {
+ fprintf(stderr, "--TGSI--------------------------------------------------------\n");
+ tgsi_dump(sel->tokens, 0);
+ }
+
+ if (rscreen->b.debug_flags & DBG_NIR) {
+ fprintf(stderr, "--NIR --------------------------------------------------------\n");
+ nir_print_shader(sel->nir, stderr);
+ }
+
+ R600_ERR("translation from NIR failed !\n");
+ goto error;
+ }
+ }
+
if (dump) {
- fprintf(stderr, "--------------------------------------------------------------\n");
- tgsi_dump(sel->tokens, 0);
-
+ if (sel->ir_type == PIPE_SHADER_IR_TGSI) {
+ fprintf(stderr, "--TGSI--------------------------------------------------------\n");
+ tgsi_dump(sel->tokens, 0);
+ }
+
if (sel->so.num_outputs) {
r600_dump_streamout(&sel->so);
}
}
- r = r600_shader_from_tgsi(rctx, shader, key);
- if (r) {
- R600_ERR("translation from TGSI failed !\n");
- goto error;
- }
+
if (shader->shader.processor_type == PIPE_SHADER_VERTEX) {
/* only disable for vertex shaders in tess paths */
if (key.vs.as_ls)
r600_bytecode_disasm(&shader->shader.bc);
fprintf(stderr, "______________________________________________________________\n");
} else if ((dump && sb_disasm) || use_sb) {
- r = r600_sb_bytecode_process(rctx, &shader->shader.bc, &shader->shader,
+ r = r600_sb_bytecode_process(rctx, &shader->shader.bc, &shader->shader,
dump, use_sb);
if (r) {
R600_ERR("r600_sb_bytecode_process failed !\n");
}
}
+ if (dump) {
+ FILE *f;
+ char fname[1024];
+ snprintf(fname, 1024, "shader_from_%s_%d.cpp",
+ (sel->ir_type == PIPE_SHADER_IR_TGSI ?
+ (rscreen->b.debug_flags & DBG_NIR ? "tgsi-nir" : "tgsi")
+ : "nir"), nshader);
+ f = fopen(fname, "w");
+ print_shader_info(f, nshader++, &shader->shader);
+ print_shader_info(stderr, nshader++, &shader->shader);
+ print_pipe_info(stderr, &sel->info);
+ if (sel->ir_type == PIPE_SHADER_IR_TGSI) {
+ fprintf(f, "/****TGSI**********************************\n");
+ tgsi_dump_to_file(sel->tokens, 0, f);
+ }
+
+ if (rscreen->b.debug_flags & DBG_NIR){
+ fprintf(f, "/****NIR **********************************\n");
+ nir_print_shader(sel->nir, f);
+ }
+ fprintf(f, "******************************************/\n");
+ fclose(f);
+ }
+
if (shader->gs_copy_shader) {
if (dump) {
// dump copy shader
void r600_pipe_shader_destroy(struct pipe_context *ctx UNUSED, struct r600_pipe_shader *shader)
{
r600_resource_reference(&shader->bo, NULL);
- r600_bytecode_clear(&shader->shader.bc);
+ if (shader->shader.bc.cf.next)
+ r600_bytecode_clear(&shader->shader.bc);
r600_release_command_buffer(&shader->command_buffer);
}
#endif
for (j = 0; j < i->Instruction.NumSrcRegs; j++) {
if (i->Src[j].Register.Dimension) {
- switch (i->Src[j].Register.File) {
- case TGSI_FILE_CONSTANT:
- case TGSI_FILE_HW_ATOMIC:
- break;
- case TGSI_FILE_INPUT:
- if (ctx->type == PIPE_SHADER_GEOMETRY ||
- ctx->type == PIPE_SHADER_TESS_CTRL ||
- ctx->type == PIPE_SHADER_TESS_EVAL)
- break;
- case TGSI_FILE_OUTPUT:
- if (ctx->type == PIPE_SHADER_TESS_CTRL)
- break;
- default:
- R600_ERR("unsupported src %d (file %d, dimension %d)\n", j,
- i->Src[j].Register.File,
- i->Src[j].Register.Dimension);
- return -EINVAL;
- }
+ switch (i->Src[j].Register.File) {
+ case TGSI_FILE_CONSTANT:
+ case TGSI_FILE_HW_ATOMIC:
+ break;
+ case TGSI_FILE_INPUT:
+ if (ctx->type == PIPE_SHADER_GEOMETRY ||
+ ctx->type == PIPE_SHADER_TESS_CTRL ||
+ ctx->type == PIPE_SHADER_TESS_EVAL)
+ break;
+ /* fallthrough */
+ case TGSI_FILE_OUTPUT:
+ if (ctx->type == PIPE_SHADER_TESS_CTRL)
+ break;
+ /* fallthrough */
+ default:
+ R600_ERR("unsupported src %d (file %d, dimension %d)\n", j,
+ i->Src[j].Register.File,
+ i->Src[j].Register.Dimension);
+ return -EINVAL;
+ }
}
}
for (j = 0; j < i->Instruction.NumDstRegs; j++) {
else {
if (name == TGSI_SEMANTIC_GENERIC) {
/* For generic params simply use sid from tgsi */
+ index = 9 + io->sid;
+ } else if (name == TGSI_SEMANTIC_TEXCOORD) {
index = io->sid;
} else {
/* For non-generic params - pack name and sid into 8 bits */
case TGSI_SEMANTIC_CLIPDIST:
assert(index <= 1);
return 2 + index;
+ case TGSI_SEMANTIC_TEXCOORD:
+ return 4 + index;
case TGSI_SEMANTIC_GENERIC:
if (index <= 63-4)
- return 4 + index - 9;
+ return 4 + index;
else
/* same explanation as in the default statement,
* the only user hitting this is st/nine.
tgsi_parse_free(&parse);
if (ctx->info.reads_samplemask &&
- (ctx->info.uses_linear_sample || ctx->info.uses_linear_sample)) {
+ (ctx->info.uses_linear_sample || ctx->info.uses_persp_sample)) {
inputs[1].enabled = true;
}
} else if (ctx->info.system_value_semantic_name[tgsi_src->Register.Index] == TGSI_SEMANTIC_TESSOUTER) {
r600_src->sel = 2;
} else if (ctx->info.system_value_semantic_name[tgsi_src->Register.Index] == TGSI_SEMANTIC_VERTICESIN) {
- if (ctx->type == PIPE_SHADER_TESS_CTRL) {
- r600_src->sel = ctx->tess_input_info;
- r600_src->swizzle[0] = 2;
- r600_src->swizzle[1] = 2;
- r600_src->swizzle[2] = 2;
- r600_src->swizzle[3] = 2;
- } else {
- r600_src->sel = ctx->tess_input_info;
- r600_src->swizzle[0] = 3;
- r600_src->swizzle[1] = 3;
- r600_src->swizzle[2] = 3;
- r600_src->swizzle[3] = 3;
- }
+ r600_src->sel = ctx->tess_input_info;
+ r600_src->swizzle[0] = 2;
+ r600_src->swizzle[1] = 2;
+ r600_src->swizzle[2] = 2;
+ r600_src->swizzle[3] = 2;
} else if (ctx->type == PIPE_SHADER_TESS_CTRL && ctx->info.system_value_semantic_name[tgsi_src->Register.Index] == TGSI_SEMANTIC_PRIMID) {
r600_src->sel = 0;
r600_src->swizzle[0] = 0;
r600_bytecode_add_alu(ctx->bc, &alu);
}
-static int generate_gs_copy_shader(struct r600_context *rctx,
- struct r600_pipe_shader *gs,
- struct pipe_stream_output_info *so)
+int generate_gs_copy_shader(struct r600_context *rctx,
+ struct r600_pipe_shader *gs,
+ struct pipe_stream_output_info *so)
{
struct r600_shader_ctx ctx = {};
struct r600_shader *gs_shader = &gs->shader;
for (i = 0; i < ctx->shader->noutput; i++) {
struct r600_bytecode_alu alu;
- int param = r600_get_lds_unique_index(ctx->shader->output[i].name, ctx->shader->output[i].sid);
+ int param = r600_get_lds_unique_index(ctx->shader->output[i].name,
+ ctx->shader->output[i].sid);
if (param) {
r = single_alu_op2(ctx, ALU_OP2_ADD_INT,
ctx.tess_input_info = ++regno;
ctx.tess_output_info = ++regno;
} else if (ctx.type == PIPE_SHADER_TESS_EVAL) {
- ctx.tess_input_info = 0;
+ ctx.tess_input_info = ++regno;
ctx.tess_output_info = ++regno;
} else if (ctx.type == PIPE_SHADER_GEOMETRY) {
ctx.gs_export_gpr_tregs[0] = ++regno;
ctx.inst_info = &eg_shader_tgsi_instruction[opcode];
else
ctx.inst_info = &r600_shader_tgsi_instruction[opcode];
+
+ ctx.bc->precise |= ctx.parse.FullToken.FullInstruction.Instruction.Precise;
+
r = ctx.inst_info->process(&ctx);
if (r)
goto out_err;
if (spilled) {
struct r600_bytecode_output cf;
- int reg = r600_get_temp(ctx);
+ int reg = 0;
int r;
+ bool add_pending_output = true;
+
+ memset(&cf, 0, sizeof(struct r600_bytecode_output));
+ get_spilled_array_base_and_size(ctx, tgsi_dst->Register.Index,
+ &cf.array_base, &cf.array_size);
+
+ /* If no component has spilled, reserve a register and add the spill code
+ * ctx->bc->n_pending_outputs is cleared after each instruction group */
+ if (ctx->bc->n_pending_outputs == 0) {
+ reg = r600_get_temp(ctx);
+ } else {
+ /* If we are already spilling and the output address is the same like
+ * before then just reuse the same slot */
+ struct r600_bytecode_output *tmpl = &ctx->bc->pending_outputs[ctx->bc->n_pending_outputs-1];
+ if ((cf.array_base + idx == tmpl->array_base) ||
+ (cf.array_base == tmpl->array_base &&
+ tmpl->index_gpr == ctx->bc->ar_reg &&
+ tgsi_dst->Register.Indirect)) {
+ reg = ctx->bc->pending_outputs[0].gpr;
+ add_pending_output = false;
+ } else {
+ reg = r600_get_temp(ctx);
+ }
+ }
r600_dst->sel = reg;
r600_dst->chan = swizzle;
r600_dst->clamp = 1;
}
- // needs to be added after op using tgsi_dst
- memset(&cf, 0, sizeof(struct r600_bytecode_output));
- cf.op = CF_OP_MEM_SCRATCH;
- cf.elem_size = 3;
- cf.gpr = reg;
- cf.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE;
- cf.mark = 1;
- cf.comp_mask = inst->Dst[0].Register.WriteMask;
- cf.swizzle_x = 0;
- cf.swizzle_y = 1;
- cf.swizzle_z = 2;
- cf.swizzle_w = 3;
- cf.burst_count = 1;
-
- get_spilled_array_base_and_size(ctx, tgsi_dst->Register.Index,
- &cf.array_base, &cf.array_size);
+ /* Add new outputs as pending */
+ if (add_pending_output) {
+ cf.op = CF_OP_MEM_SCRATCH;
+ cf.elem_size = 3;
+ cf.gpr = reg;
+ cf.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE;
+ cf.mark = 1;
+ cf.comp_mask = inst->Dst[0].Register.WriteMask;
+ cf.swizzle_x = 0;
+ cf.swizzle_y = 1;
+ cf.swizzle_z = 2;
+ cf.swizzle_w = 3;
+ cf.burst_count = 1;
- if (tgsi_dst->Register.Indirect) {
- if (ctx->bc->chip_class < R700)
- cf.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE_IND;
- else
- cf.type = 3; // V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE_IND_ACK;
- cf.index_gpr = ctx->bc->ar_reg;
+ if (tgsi_dst->Register.Indirect) {
+ if (ctx->bc->chip_class < R700)
+ cf.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE_IND;
+ else
+ cf.type = 3; // V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE_IND_ACK;
+ cf.index_gpr = ctx->bc->ar_reg;
}
else {
cf.array_base += idx;
if (ctx->bc->chip_class >= R700)
r600_bytecode_need_wait_ack(ctx->bc, true);
-
+ }
return;
}
else {
struct r600_bytecode_alu alu;
int i, r, j;
unsigned write_mask = inst->Dst[0].Register.WriteMask;
+ int lasti = tgsi_last_instruction(write_mask);
int tmp0 = ctx->temp_reg;
int tmp1 = r600_get_temp(ctx);
int tmp2 = r600_get_temp(ctx);
int tmp3 = r600_get_temp(ctx);
+ int tmp4 = 0;
+
+ /* Use additional temp if dst register and src register are the same */
+ if (inst->Src[0].Register.Index == inst->Dst[0].Register.Index ||
+ inst->Src[1].Register.Index == inst->Dst[0].Register.Index) {
+ tmp4 = r600_get_temp(ctx);
+ }
+
/* Unsigned path:
*
* we need to represent src1 as src2*q + r, where q - quotient, r - remainder
alu.dst.chan = 2;
alu.dst.write = 1;
} else {
- tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ if (tmp4 > 0) {
+ alu.dst.sel = tmp4;
+ alu.dst.chan = i;
+ alu.dst.write = 1;
+ } else {
+ tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ }
}
alu.src[0].sel = tmp1;
alu.op = ALU_OP3_CNDGE_INT;
alu.is_op3 = 1;
- tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ if (tmp4 > 0) {
+ alu.dst.sel = tmp4;
+ alu.dst.chan = i;
+ alu.dst.write = 1;
+ } else {
+ tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ }
r600_bytecode_src(&alu.src[0], &ctx->src[0], i);
alu.src[1].sel = tmp0;
alu.op = ALU_OP3_CNDGE_INT;
alu.is_op3 = 1;
- tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ if (tmp4 > 0) {
+ alu.dst.sel = tmp4;
+ alu.dst.chan = i;
+ alu.dst.write = 1;
+ } else {
+ tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ }
alu.src[0].sel = tmp2;
alu.src[0].chan = 2;
}
}
}
+
+ if (tmp4 > 0) {
+ for (i = 0; i <= lasti; ++i) {
+ if (!(write_mask & (1<<i)))
+ continue;
+
+ memset(&alu, 0, sizeof(struct r600_bytecode_alu));
+ alu.op = ALU_OP1_MOV;
+ tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ alu.src[0].sel = tmp4;
+ alu.src[0].chan = i;
+
+ if (i == lasti)
+ alu.last = 1;
+ if ((r = r600_bytecode_add_alu(ctx->bc, &alu)))
+ return r;
+ }
+ }
+
return 0;
}
static int tgsi_ssg(struct r600_shader_ctx *ctx)
{
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
+ unsigned write_mask = inst->Dst[0].Register.WriteMask;
+ int last_inst = tgsi_last_instruction(write_mask);
struct r600_bytecode_alu alu;
int i, r;
/* tmp = (src > 0 ? 1 : src) */
- for (i = 0; i < 4; i++) {
+ for (i = 0; i <= last_inst; i++) {
+ if (!(write_mask & (1 << i)))
+ continue;
memset(&alu, 0, sizeof(struct r600_bytecode_alu));
alu.op = ALU_OP3_CNDGT;
alu.is_op3 = 1;
alu.src[1].sel = V_SQ_ALU_SRC_1;
r600_bytecode_src(&alu.src[2], &ctx->src[0], i);
- if (i == 3)
+ if (i == last_inst)
alu.last = 1;
r = r600_bytecode_add_alu(ctx->bc, &alu);
if (r)
}
/* dst = (-tmp > 0 ? -1 : tmp) */
- for (i = 0; i < 4; i++) {
+ for (i = 0; i <= last_inst; i++) {
+ if (!(write_mask & (1 << i)))
+ continue;
memset(&alu, 0, sizeof(struct r600_bytecode_alu));
alu.op = ALU_OP3_CNDGT;
alu.is_op3 = 1;
alu.src[2].sel = ctx->temp_reg;
alu.src[2].chan = i;
- if (i == 3)
+ if (i == last_inst)
alu.last = 1;
r = r600_bytecode_add_alu(ctx->bc, &alu);
if (r)
}
else {
location = TGSI_INTERPOLATE_LOC_CENTROID;
+ ctx->shader->input[input].uses_interpolate_at_centroid = 1;
}
k = eg_get_interpolator_index(ctx->shader->input[input].interpolate, location);
}
static int tgsi_make_src_for_op3(struct r600_shader_ctx *ctx,
- unsigned temp, int chan,
+ unsigned writemask,
struct r600_bytecode_alu_src *bc_src,
const struct r600_shader_src *shader_src)
{
struct r600_bytecode_alu alu;
- int r;
+ int i, r;
+ int lasti = tgsi_last_instruction(writemask);
+ int temp_reg = 0;
- r600_bytecode_src(bc_src, shader_src, chan);
+ r600_bytecode_src(&bc_src[0], shader_src, 0);
+ r600_bytecode_src(&bc_src[1], shader_src, 1);
+ r600_bytecode_src(&bc_src[2], shader_src, 2);
+ r600_bytecode_src(&bc_src[3], shader_src, 3);
- /* op3 operands don't support abs modifier */
if (bc_src->abs) {
- assert(temp!=0); /* we actually need the extra register, make sure it is allocated. */
- memset(&alu, 0, sizeof(struct r600_bytecode_alu));
- alu.op = ALU_OP1_MOV;
- alu.dst.sel = temp;
- alu.dst.chan = chan;
- alu.dst.write = 1;
+ temp_reg = r600_get_temp(ctx);
- alu.src[0] = *bc_src;
- alu.last = true; // sufficient?
- r = r600_bytecode_add_alu(ctx->bc, &alu);
- if (r)
- return r;
-
- memset(bc_src, 0, sizeof(*bc_src));
- bc_src->sel = temp;
- bc_src->chan = chan;
+ for (i = 0; i < lasti + 1; i++) {
+ if (!(writemask & (1 << i)))
+ continue;
+ memset(&alu, 0, sizeof(struct r600_bytecode_alu));
+ alu.op = ALU_OP1_MOV;
+ alu.dst.sel = temp_reg;
+ alu.dst.chan = i;
+ alu.dst.write = 1;
+ alu.src[0] = bc_src[i];
+ if (i == lasti) {
+ alu.last = 1;
+ }
+ r = r600_bytecode_add_alu(ctx->bc, &alu);
+ if (r)
+ return r;
+ memset(&bc_src[i], 0, sizeof(*bc_src));
+ bc_src[i].sel = temp_reg;
+ bc_src[i].chan = i;
+ }
}
return 0;
}
{
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
struct r600_bytecode_alu alu;
+ struct r600_bytecode_alu_src srcs[4][4];
int i, j, r;
int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
- int temp_regs[4];
unsigned op = ctx->inst_info->op;
if (op == ALU_OP3_MULADD_IEEE &&
op = ALU_OP3_MULADD;
for (j = 0; j < inst->Instruction.NumSrcRegs; j++) {
- temp_regs[j] = 0;
- if (ctx->src[j].abs)
- temp_regs[j] = r600_get_temp(ctx);
+ r = tgsi_make_src_for_op3(ctx, inst->Dst[0].Register.WriteMask,
+ srcs[j], &ctx->src[j]);
+ if (r)
+ return r;
}
+
for (i = 0; i < lasti + 1; i++) {
if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
continue;
memset(&alu, 0, sizeof(struct r600_bytecode_alu));
alu.op = op;
for (j = 0; j < inst->Instruction.NumSrcRegs; j++) {
- r = tgsi_make_src_for_op3(ctx, temp_regs[j], i, &alu.src[j], &ctx->src[j]);
- if (r)
- return r;
+ alu.src[j] = srcs[j][i];
}
if (dst == -1) {
{
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
struct r600_bytecode_tex tex;
+ struct r600_bytecode_tex grad_offs[3];
struct r600_bytecode_alu alu;
unsigned src_gpr;
- int r, i, j;
+ int r, i, j, n_grad_offs = 0;
int opcode;
bool read_compressed_msaa = ctx->bc->has_compressed_msaa_texturing &&
inst->Instruction.Opcode == TGSI_OPCODE_TXF &&
int8_t offset_x = 0, offset_y = 0, offset_z = 0;
boolean has_txq_cube_array_z = false;
unsigned sampler_index_mode;
+ int array_index_offset_channel = -1;
if (inst->Instruction.Opcode == TGSI_OPCODE_TXQ &&
((inst->Texture.Texture == TGSI_TEXTURE_CUBE_ARRAY ||
if (r)
return r;
+ /* Evaluate the array index according to floor(idx + 0.5). This
+ * needs to be done before merging the face select value, because
+ * otherwise the fractional part of the array index will interfere
+ * with the face select value */
+ memset(&alu, 0, sizeof(struct r600_bytecode_alu));
+ r600_bytecode_src(&alu.src[0], &ctx->src[0], 3);
+ alu.op = ALU_OP1_RNDNE;
+ alu.dst.sel = ctx->temp_reg;
+ alu.dst.chan = 3;
+ alu.dst.write = 1;
+ alu.last = 1;
+ r = r600_bytecode_add_alu(ctx->bc, &alu);
+ if (r)
+ return r;
+
+ /* Because the array slice index and the cube face index are merged
+ * into one value we have to make sure the array slice index is >= 0,
+ * otherwise the face selection will fail */
+ memset(&alu, 0, sizeof(struct r600_bytecode_alu));
+ alu.op = ALU_OP2_MAX;
+ alu.src[0].sel = ctx->temp_reg;
+ alu.src[0].chan = 3;
+ alu.src[1].sel = V_SQ_ALU_SRC_0;
+ alu.dst.sel = ctx->temp_reg;
+ alu.dst.chan = 3;
+ alu.dst.write = 1;
+ alu.last = 1;
+ r = r600_bytecode_add_alu(ctx->bc, &alu);
+ if (r)
+ return r;
+
/* have to multiply original layer by 8 and add to face id (temp.w) in Z */
memset(&alu, 0, sizeof(struct r600_bytecode_alu));
alu.op = ALU_OP3_MULADD;
alu.is_op3 = 1;
- r600_bytecode_src(&alu.src[0], &ctx->src[0], 3);
+ alu.src[0].sel = ctx->temp_reg;
+ alu.src[0].chan = 3;
alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
alu.src[1].chan = 0;
alu.src[1].value = u_bitcast_f2u(8.0f);
}
for (i = 1; i < 3; i++) {
/* set gradients h/v */
- memset(&tex, 0, sizeof(struct r600_bytecode_tex));
- tex.op = (i == 1) ? FETCH_OP_SET_GRADIENTS_H :
+ struct r600_bytecode_tex *t = &grad_offs[n_grad_offs++];
+ memset(t, 0, sizeof(struct r600_bytecode_tex));
+ t->op = (i == 1) ? FETCH_OP_SET_GRADIENTS_H :
FETCH_OP_SET_GRADIENTS_V;
- tex.sampler_id = tgsi_tex_get_src_gpr(ctx, sampler_src_reg);
- tex.sampler_index_mode = sampler_index_mode;
- tex.resource_id = tex.sampler_id + R600_MAX_CONST_BUFFERS;
- tex.resource_index_mode = sampler_index_mode;
-
- tex.src_gpr = (i == 1) ? temp_h : temp_v;
- tex.src_sel_x = 0;
- tex.src_sel_y = 1;
- tex.src_sel_z = 2;
- tex.src_sel_w = 3;
-
- tex.dst_gpr = r600_get_temp(ctx); /* just to avoid confusing the asm scheduler */
- tex.dst_sel_x = tex.dst_sel_y = tex.dst_sel_z = tex.dst_sel_w = 7;
+ t->sampler_id = tgsi_tex_get_src_gpr(ctx, sampler_src_reg);
+ t->sampler_index_mode = sampler_index_mode;
+ t->resource_id = t->sampler_id + R600_MAX_CONST_BUFFERS;
+ t->resource_index_mode = sampler_index_mode;
+
+ t->src_gpr = (i == 1) ? temp_h : temp_v;
+ t->src_sel_x = 0;
+ t->src_sel_y = 1;
+ t->src_sel_z = 2;
+ t->src_sel_w = 3;
+
+ t->dst_gpr = r600_get_temp(ctx); /* just to avoid confusing the asm scheduler */
+ t->dst_sel_x = t->dst_sel_y = t->dst_sel_z = t->dst_sel_w = 7;
if (inst->Texture.Texture != TGSI_TEXTURE_RECT) {
- tex.coord_type_x = 1;
- tex.coord_type_y = 1;
- tex.coord_type_z = 1;
- tex.coord_type_w = 1;
+ t->coord_type_x = 1;
+ t->coord_type_y = 1;
+ t->coord_type_z = 1;
+ t->coord_type_w = 1;
}
- r = r600_bytecode_add_tex(ctx->bc, &tex);
- if (r)
- return r;
}
}
if (opcode == FETCH_OP_GATHER4 &&
inst->TexOffsets[0].File != TGSI_FILE_NULL &&
inst->TexOffsets[0].File != TGSI_FILE_IMMEDIATE) {
+ struct r600_bytecode_tex *t;
opcode = FETCH_OP_GATHER4_O;
/* GATHER4_O/GATHER4_C_O use offset values loaded by
SET_TEXTURE_OFFSETS instruction. The immediate offset values
encoded in the instruction are ignored. */
- memset(&tex, 0, sizeof(struct r600_bytecode_tex));
- tex.op = FETCH_OP_SET_TEXTURE_OFFSETS;
- tex.sampler_id = tgsi_tex_get_src_gpr(ctx, sampler_src_reg);
- tex.sampler_index_mode = sampler_index_mode;
- tex.resource_id = tex.sampler_id + R600_MAX_CONST_BUFFERS;
- tex.resource_index_mode = sampler_index_mode;
-
- tex.src_gpr = ctx->file_offset[inst->TexOffsets[0].File] + inst->TexOffsets[0].Index;
- tex.src_sel_x = inst->TexOffsets[0].SwizzleX;
- tex.src_sel_y = inst->TexOffsets[0].SwizzleY;
- tex.src_sel_z = inst->TexOffsets[0].SwizzleZ;
- tex.src_sel_w = 4;
+ t = &grad_offs[n_grad_offs++];
+ memset(t, 0, sizeof(struct r600_bytecode_tex));
+ t->op = FETCH_OP_SET_TEXTURE_OFFSETS;
+ t->sampler_id = tgsi_tex_get_src_gpr(ctx, sampler_src_reg);
+ t->sampler_index_mode = sampler_index_mode;
+ t->resource_id = t->sampler_id + R600_MAX_CONST_BUFFERS;
+ t->resource_index_mode = sampler_index_mode;
+
+ t->src_gpr = ctx->file_offset[inst->TexOffsets[0].File] + inst->TexOffsets[0].Index;
+ t->src_sel_x = inst->TexOffsets[0].SwizzleX;
+ t->src_sel_y = inst->TexOffsets[0].SwizzleY;
+ if (inst->Texture.Texture == TGSI_TEXTURE_2D_ARRAY ||
+ inst->Texture.Texture == TGSI_TEXTURE_SHADOW2D_ARRAY)
+ /* make sure array index selector is 0, this is just a safety
+ * precausion because TGSI seems to emit something strange here */
+ t->src_sel_z = 4;
+ else
+ t->src_sel_z = inst->TexOffsets[0].SwizzleZ;
- tex.dst_sel_x = 7;
- tex.dst_sel_y = 7;
- tex.dst_sel_z = 7;
- tex.dst_sel_w = 7;
+ t->src_sel_w = 4;
- r = r600_bytecode_add_tex(ctx->bc, &tex);
- if (r)
- return r;
+ t->dst_sel_x = 7;
+ t->dst_sel_y = 7;
+ t->dst_sel_z = 7;
+ t->dst_sel_w = 7;
}
if (inst->Texture.Texture == TGSI_TEXTURE_SHADOW1D ||
opcode == FETCH_OP_SAMPLE_C_LB) {
/* the array index is read from Y */
tex.coord_type_y = 0;
+ array_index_offset_channel = tex.src_sel_y;
} else {
/* the array index is read from Z */
tex.coord_type_z = 0;
tex.src_sel_z = tex.src_sel_y;
+ array_index_offset_channel = tex.src_sel_z;
}
} else if (inst->Texture.Texture == TGSI_TEXTURE_2D_ARRAY ||
- inst->Texture.Texture == TGSI_TEXTURE_SHADOW2D_ARRAY ||
- ((inst->Texture.Texture == TGSI_TEXTURE_CUBE_ARRAY ||
+ inst->Texture.Texture == TGSI_TEXTURE_SHADOW2D_ARRAY) {
+ tex.coord_type_z = 0;
+ array_index_offset_channel = tex.src_sel_z;
+ } else if ((inst->Texture.Texture == TGSI_TEXTURE_CUBE_ARRAY ||
inst->Texture.Texture == TGSI_TEXTURE_SHADOWCUBE_ARRAY) &&
- (ctx->bc->chip_class >= EVERGREEN)))
- /* the array index is read from Z */
+ (ctx->bc->chip_class >= EVERGREEN))
+ /* the array index is read from Z, coordinate will be corrected elsewhere */
tex.coord_type_z = 0;
+ /* We have array access to 1D or 2D ARRAY, the coordinates are not int ->
+ * evaluate the array index */
+ if (array_index_offset_channel >= 0 &&
+ opcode != FETCH_OP_LD &&
+ opcode != FETCH_OP_GET_TEXTURE_RESINFO) {
+ memset(&alu, 0, sizeof(struct r600_bytecode_alu));
+ alu.src[0].sel = tex.src_gpr;
+ alu.src[0].chan = array_index_offset_channel;
+ alu.src[0].rel = tex.src_rel;
+ alu.op = ALU_OP1_RNDNE;
+ alu.dst.sel = tex.src_gpr;
+ alu.dst.chan = array_index_offset_channel;
+ alu.dst.rel = tex.src_rel;
+ alu.dst.write = 1;
+ alu.last = 1;
+ r = r600_bytecode_add_alu(ctx->bc, &alu);
+ if (r)
+ return r;
+ }
+
/* mask unused source components */
if (opcode == FETCH_OP_SAMPLE || opcode == FETCH_OP_GATHER4) {
switch (inst->Texture.Texture) {
}
}
+ /* Emit set gradient and offset instructions. */
+ for (i = 0; i < n_grad_offs; ++i) {
+ r = r600_bytecode_add_tex(ctx->bc, &grad_offs[i]);
+ if (r)
+ return r;
+ }
+
r = r600_bytecode_add_tex(ctx->bc, &tex);
if (r)
return r;
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
struct r600_bytecode_alu alu;
unsigned lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
- unsigned i, temp_regs[2];
+ struct r600_bytecode_alu_src srcs[2][4];
+ unsigned i;
int r;
/* optimize if it's just an equal balance */
}
/* src0 * src1 + (1 - src0) * src2 */
- if (ctx->src[0].abs)
- temp_regs[0] = r600_get_temp(ctx);
- else
- temp_regs[0] = 0;
- if (ctx->src[1].abs)
- temp_regs[1] = r600_get_temp(ctx);
- else
- temp_regs[1] = 0;
+
+ for (i = 0; i < 2; i++) {
+ r = tgsi_make_src_for_op3(ctx, inst->Dst[0].Register.WriteMask,
+ srcs[i], &ctx->src[i]);
+ if (r)
+ return r;
+ }
for (i = 0; i < lasti + 1; i++) {
if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
memset(&alu, 0, sizeof(struct r600_bytecode_alu));
alu.op = ALU_OP3_MULADD;
alu.is_op3 = 1;
- r = tgsi_make_src_for_op3(ctx, temp_regs[0], i, &alu.src[0], &ctx->src[0]);
- if (r)
- return r;
- r = tgsi_make_src_for_op3(ctx, temp_regs[1], i, &alu.src[1], &ctx->src[1]);
- if (r)
- return r;
+ alu.src[0] = srcs[0][i];
+ alu.src[1] = srcs[1][i];
alu.src[2].sel = ctx->temp_reg;
alu.src[2].chan = i;
struct r600_bytecode_alu alu;
int i, r, j;
int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
- int temp_regs[3];
+ struct r600_bytecode_alu_src srcs[3][4];
+
unsigned op;
if (ctx->src[0].abs && ctx->src[0].neg) {
}
for (j = 0; j < inst->Instruction.NumSrcRegs; j++) {
- temp_regs[j] = 0;
- if (ctx->src[j].abs)
- temp_regs[j] = r600_get_temp(ctx);
+ r = tgsi_make_src_for_op3(ctx, inst->Dst[0].Register.WriteMask,
+ srcs[j], &ctx->src[j]);
+ if (r)
+ return r;
}
for (i = 0; i < lasti + 1; i++) {
memset(&alu, 0, sizeof(struct r600_bytecode_alu));
alu.op = op;
- r = tgsi_make_src_for_op3(ctx, temp_regs[0], i, &alu.src[0], &ctx->src[0]);
- if (r)
- return r;
- r = tgsi_make_src_for_op3(ctx, temp_regs[2], i, &alu.src[1], &ctx->src[2]);
- if (r)
- return r;
- r = tgsi_make_src_for_op3(ctx, temp_regs[1], i, &alu.src[2], &ctx->src[1]);
- if (r)
- return r;
+ alu.src[0] = srcs[0][i];
+ alu.src[1] = srcs[2][i];
+ alu.src[2] = srcs[1][i];
+
tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
alu.dst.chan = i;
alu.dst.write = 1;
projects / mesa.git / blobdiff