From feca6609390d4642418cf7aab878e654964510c4 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Mon, 4 Oct 2010 15:08:03 -0700 Subject: [PATCH] i965: Fix up IF/ELSE/ENDIF for gen6. The jump delta is now in the part of the instruction where the destination fields used to be, and the src args are ignored (or not, for the new non-predicated IF that we don't use yet). --- src/mesa/drivers/dri/i965/brw_disasm.c | 5 ++ src/mesa/drivers/dri/i965/brw_eu_emit.c | 86 ++++++++++++++++++------- src/mesa/drivers/dri/i965/brw_structs.h | 12 ++++ 3 files changed, 79 insertions(+), 24 deletions(-) diff --git a/src/mesa/drivers/dri/i965/brw_disasm.c b/src/mesa/drivers/dri/i965/brw_disasm.c index 3a93b253777..be27f9226c3 100644 --- a/src/mesa/drivers/dri/i965/brw_disasm.c +++ b/src/mesa/drivers/dri/i965/brw_disasm.c @@ -892,7 +892,12 @@ int brw_disasm (FILE *file, struct brw_instruction *inst, int gen) if (opcode[inst->header.opcode].ndst > 0) { pad (file, 16); err |= dest (file, inst); + } else if (gen >= 6 && (inst->header.opcode == BRW_OPCODE_IF || + inst->header.opcode == BRW_OPCODE_ELSE || + inst->header.opcode == BRW_OPCODE_ENDIF)) { + format (file, " %d", inst->bits1.branch_gen6.jump_count); } + if (opcode[inst->header.opcode].nsrc > 0) { pad (file, 32); err |= src0 (file, inst); diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c index fbc6894d054..419b40ba84b 100644 --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c @@ -787,6 +787,7 @@ struct brw_instruction *brw_JMPI(struct brw_compile *p, */ struct brw_instruction *brw_IF(struct brw_compile *p, GLuint execute_size) { + struct intel_context *intel = &p->brw->intel; struct brw_instruction *insn; if (p->single_program_flow) { @@ -800,9 +801,15 @@ struct brw_instruction *brw_IF(struct brw_compile *p, GLuint execute_size) /* Override the defaults for this instruction: */ - brw_set_dest(insn, brw_ip_reg()); - brw_set_src0(insn, brw_ip_reg()); - brw_set_src1(insn, brw_imm_d(0x0)); + if (intel->gen < 6) { + brw_set_dest(insn, brw_ip_reg()); + brw_set_src0(insn, brw_ip_reg()); + brw_set_src1(insn, brw_imm_d(0x0)); + } else { + brw_set_dest(insn, brw_imm_w(0)); + brw_set_src0(insn, brw_null_reg()); + brw_set_src1(insn, brw_null_reg()); + } insn->header.execution_size = execute_size; insn->header.compression_control = BRW_COMPRESSION_NONE; @@ -835,9 +842,15 @@ struct brw_instruction *brw_ELSE(struct brw_compile *p, insn = next_insn(p, BRW_OPCODE_ELSE); } - brw_set_dest(insn, brw_ip_reg()); - brw_set_src0(insn, brw_ip_reg()); - brw_set_src1(insn, brw_imm_d(0x0)); + if (intel->gen < 6) { + brw_set_dest(insn, brw_ip_reg()); + brw_set_src0(insn, brw_ip_reg()); + brw_set_src1(insn, brw_imm_d(0x0)); + } else { + brw_set_dest(insn, brw_imm_w(0)); + brw_set_src0(insn, brw_null_reg()); + brw_set_src1(insn, brw_null_reg()); + } insn->header.compression_control = BRW_COMPRESSION_NONE; insn->header.execution_size = if_insn->header.execution_size; @@ -854,9 +867,13 @@ struct brw_instruction *brw_ELSE(struct brw_compile *p, } else { assert(if_insn->header.opcode == BRW_OPCODE_IF); - if_insn->bits3.if_else.jump_count = br * (insn - if_insn); - if_insn->bits3.if_else.pop_count = 0; - if_insn->bits3.if_else.pad0 = 0; + if (intel->gen < 6) { + if_insn->bits3.if_else.jump_count = br * (insn - if_insn); + if_insn->bits3.if_else.pop_count = 0; + if_insn->bits3.if_else.pad0 = 0; + } else { + if_insn->bits1.branch_gen6.jump_count = br * (insn - if_insn + 1); + } } return insn; @@ -884,9 +901,15 @@ void brw_ENDIF(struct brw_compile *p, } else { struct brw_instruction *insn = next_insn(p, BRW_OPCODE_ENDIF); - brw_set_dest(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD)); - brw_set_src0(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD)); - brw_set_src1(insn, brw_imm_d(0x0)); + if (intel->gen < 6) { + brw_set_dest(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD)); + brw_set_src0(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD)); + brw_set_src1(insn, brw_imm_d(0x0)); + } else { + brw_set_dest(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_W)); + brw_set_src0(insn, brw_null_reg()); + brw_set_src1(insn, brw_null_reg()); + } insn->header.compression_control = BRW_COMPRESSION_NONE; insn->header.execution_size = patch_insn->header.execution_size; @@ -900,26 +923,41 @@ void brw_ENDIF(struct brw_compile *p, */ if (patch_insn->header.opcode == BRW_OPCODE_IF) { if (intel->gen < 6) { - /* Automagically turn it into an IFF: + /* Turn it into an IFF, which means no mask stack operations for + * all-false and jumping past the ENDIF. */ patch_insn->header.opcode = BRW_OPCODE_IFF; + patch_insn->bits3.if_else.jump_count = br * (insn - patch_insn + 1); + patch_insn->bits3.if_else.pop_count = 0; + patch_insn->bits3.if_else.pad0 = 0; + } else { + /* As of gen6, there is no IFF and IF must point to the ENDIF. */ + patch_insn->bits1.branch_gen6.jump_count = br * (insn - patch_insn); } - patch_insn->bits3.if_else.jump_count = br * (insn - patch_insn + 1); - patch_insn->bits3.if_else.pop_count = 0; - patch_insn->bits3.if_else.pad0 = 0; - } else if (patch_insn->header.opcode == BRW_OPCODE_ELSE) { - patch_insn->bits3.if_else.jump_count = br * (insn - patch_insn + 1); - patch_insn->bits3.if_else.pop_count = 1; - patch_insn->bits3.if_else.pad0 = 0; } else { - assert(0); + assert(patch_insn->header.opcode == BRW_OPCODE_ELSE); + if (intel->gen < 6) { + /* BRW_OPCODE_ELSE pre-gen6 should point just past the + * matching ENDIF. + */ + patch_insn->bits3.if_else.jump_count = br * (insn - patch_insn + 1); + patch_insn->bits3.if_else.pop_count = 1; + patch_insn->bits3.if_else.pad0 = 0; + } else { + /* BRW_OPCODE_ELSE on gen6 should point to the matching ENDIF. */ + patch_insn->bits1.branch_gen6.jump_count = br * (insn - patch_insn); + } } /* Also pop item off the stack in the endif instruction: */ - insn->bits3.if_else.jump_count = 0; - insn->bits3.if_else.pop_count = 1; - insn->bits3.if_else.pad0 = 0; + if (intel->gen < 6) { + insn->bits3.if_else.jump_count = 0; + insn->bits3.if_else.pop_count = 1; + insn->bits3.if_else.pad0 = 0; + } else { + insn->bits1.branch_gen6.jump_count = 2; + } } } diff --git a/src/mesa/drivers/dri/i965/brw_structs.h b/src/mesa/drivers/dri/i965/brw_structs.h index 7b919872c40..8ce9af9c4fe 100644 --- a/src/mesa/drivers/dri/i965/brw_structs.h +++ b/src/mesa/drivers/dri/i965/brw_structs.h @@ -1381,6 +1381,18 @@ struct brw_instruction GLuint dest_horiz_stride:2; GLuint dest_address_mode:1; } ia16; + + struct { + GLuint dest_reg_file:2; + GLuint dest_reg_type:3; + GLuint src0_reg_file:2; + GLuint src0_reg_type:3; + GLuint src1_reg_file:2; + GLuint src1_reg_type:3; + GLuint pad:1; + + GLint jump_count:16; + } branch_gen6; } bits1; -- 2.30.2