From 2cff3242109078999c57d5e6772418c09e835826 Mon Sep 17 00:00:00 2001 From: Matt Turner Date: Wed, 14 Jun 2017 11:03:19 -0700 Subject: [PATCH] intel/compiler: Add Gen11+ native float type This new type exposes the additional precision offered by the accumulator register and will be used in the next patch to implement the functionality of the PLN instruction using a pair of MAD instructions. One weird thing to note: align1 ternary instructions may only have an accumulator in the dst or src1 normally, but when src0's type is :NF the accumulator is read. Reviewed-by: Kenneth Graunke --- src/intel/compiler/brw_disasm.c | 7 +++++++ src/intel/compiler/brw_eu_emit.c | 10 ++++++++-- src/intel/compiler/brw_eu_validate.c | 1 + src/intel/compiler/brw_reg_type.c | 8 ++++++++ src/intel/compiler/brw_reg_type.h | 2 ++ src/intel/compiler/brw_shader.cpp | 6 ++++++ 6 files changed, 32 insertions(+), 2 deletions(-) diff --git a/src/intel/compiler/brw_disasm.c b/src/intel/compiler/brw_disasm.c index 429ed781404..a9a108f8acd 100644 --- a/src/intel/compiler/brw_disasm.c +++ b/src/intel/compiler/brw_disasm.c @@ -1035,6 +1035,12 @@ src0_3src(FILE *file, const struct gen_device_info *devinfo, const brw_inst *ins reg_nr = brw_inst_3src_src0_reg_nr(devinfo, inst); subreg_nr = brw_inst_3src_a1_src0_subreg_nr(devinfo, inst); type = brw_inst_3src_a1_src0_type(devinfo, inst); + } else if (brw_inst_3src_a1_src0_type(devinfo, inst) == + BRW_REGISTER_TYPE_NF) { + _file = BRW_ARCHITECTURE_REGISTER_FILE; + reg_nr = brw_inst_3src_src0_reg_nr(devinfo, inst); + subreg_nr = brw_inst_3src_a1_src0_subreg_nr(devinfo, inst); + type = brw_inst_3src_a1_src0_type(devinfo, inst); } else { _file = BRW_IMMEDIATE_VALUE; uint16_t imm_val = brw_inst_3src_a1_src0_imm(devinfo, inst); @@ -1288,6 +1294,7 @@ imm(FILE *file, const struct gen_device_info *devinfo, enum brw_reg_type type, case BRW_REGISTER_TYPE_HF: string(file, "Half Float IMM"); break; + case BRW_REGISTER_TYPE_NF: case BRW_REGISTER_TYPE_UB: case BRW_REGISTER_TYPE_B: format(file, "*** invalid immediate type %d ", type); diff --git a/src/intel/compiler/brw_eu_emit.c b/src/intel/compiler/brw_eu_emit.c index c25d8d6eda0..ec871e5aa75 100644 --- a/src/intel/compiler/brw_eu_emit.c +++ b/src/intel/compiler/brw_eu_emit.c @@ -771,7 +771,11 @@ brw_alu3(struct brw_codegen *p, unsigned opcode, struct brw_reg dest, to_3src_align1_hstride(src2.hstride)); brw_inst_set_3src_a1_src0_subreg_nr(devinfo, inst, src0.subnr); - brw_inst_set_3src_src0_reg_nr(devinfo, inst, src0.nr); + if (src0.type == BRW_REGISTER_TYPE_NF) { + brw_inst_set_3src_src0_reg_nr(devinfo, inst, BRW_ARF_ACCUMULATOR); + } else { + brw_inst_set_3src_src0_reg_nr(devinfo, inst, src0.nr); + } brw_inst_set_3src_src0_abs(devinfo, inst, src0.abs); brw_inst_set_3src_src0_negate(devinfo, inst, src0.negate); @@ -790,7 +794,9 @@ brw_alu3(struct brw_codegen *p, unsigned opcode, struct brw_reg dest, brw_inst_set_3src_src2_negate(devinfo, inst, src2.negate); assert(src0.file == BRW_GENERAL_REGISTER_FILE || - src0.file == BRW_IMMEDIATE_VALUE); + src0.file == BRW_IMMEDIATE_VALUE || + (src0.file == BRW_ARCHITECTURE_REGISTER_FILE && + src0.type == BRW_REGISTER_TYPE_NF)); assert(src1.file == BRW_GENERAL_REGISTER_FILE || src1.file == BRW_ARCHITECTURE_REGISTER_FILE); assert(src2.file == BRW_GENERAL_REGISTER_FILE || diff --git a/src/intel/compiler/brw_eu_validate.c b/src/intel/compiler/brw_eu_validate.c index 6ee6b4ffbe7..d3189d1ef5e 100644 --- a/src/intel/compiler/brw_eu_validate.c +++ b/src/intel/compiler/brw_eu_validate.c @@ -277,6 +277,7 @@ static enum brw_reg_type execution_type_for_type(enum brw_reg_type type) { switch (type) { + case BRW_REGISTER_TYPE_NF: case BRW_REGISTER_TYPE_DF: case BRW_REGISTER_TYPE_F: case BRW_REGISTER_TYPE_HF: diff --git a/src/intel/compiler/brw_reg_type.c b/src/intel/compiler/brw_reg_type.c index c4f8eedeb4b..3c82eb0a76f 100644 --- a/src/intel/compiler/brw_reg_type.c +++ b/src/intel/compiler/brw_reg_type.c @@ -52,6 +52,7 @@ enum hw_reg_type { GEN11_HW_REG_TYPE_HF = 8, GEN11_HW_REG_TYPE_F = 9, GEN11_HW_REG_TYPE_DF = 10, + GEN11_HW_REG_TYPE_NF = 11, }; enum hw_imm_type { @@ -87,6 +88,8 @@ static const struct hw_type { enum hw_reg_type reg_type; enum hw_imm_type imm_type; } gen4_hw_type[] = { + [0 ... BRW_REGISTER_TYPE_LAST] = { INVALID, INVALID }, + [BRW_REGISTER_TYPE_DF] = { GEN7_HW_REG_TYPE_DF, GEN8_HW_IMM_TYPE_DF }, [BRW_REGISTER_TYPE_F] = { BRW_HW_REG_TYPE_F, BRW_HW_IMM_TYPE_F }, [BRW_REGISTER_TYPE_HF] = { GEN8_HW_REG_TYPE_HF, GEN8_HW_IMM_TYPE_HF }, @@ -103,6 +106,7 @@ static const struct hw_type { [BRW_REGISTER_TYPE_V] = { INVALID, BRW_HW_IMM_TYPE_V }, [BRW_REGISTER_TYPE_UV] = { INVALID, BRW_HW_IMM_TYPE_UV }, }, gen11_hw_type[] = { + [BRW_REGISTER_TYPE_NF] = { GEN11_HW_REG_TYPE_NF, INVALID }, [BRW_REGISTER_TYPE_DF] = { GEN11_HW_REG_TYPE_DF, GEN11_HW_IMM_TYPE_DF }, [BRW_REGISTER_TYPE_F] = { GEN11_HW_REG_TYPE_F, GEN11_HW_IMM_TYPE_F }, [BRW_REGISTER_TYPE_HF] = { GEN11_HW_REG_TYPE_HF, GEN11_HW_IMM_TYPE_HF }, @@ -139,6 +143,7 @@ enum hw_3src_reg_type { GEN10_ALIGN1_3SRC_REG_TYPE_HF = 0b000, GEN10_ALIGN1_3SRC_REG_TYPE_F = 0b001, GEN10_ALIGN1_3SRC_REG_TYPE_DF = 0b010, + GEN11_ALIGN1_3SRC_REG_TYPE_NF = 0b011, /** @} */ /** When ExecutionDatatype is 0: @{ */ @@ -165,6 +170,7 @@ static const struct hw_3src_type { #define E(x) BRW_ALIGN1_3SRC_EXEC_TYPE_##x [0 ... BRW_REGISTER_TYPE_LAST] = { INVALID }, + [BRW_REGISTER_TYPE_NF] = { GEN11_ALIGN1_3SRC_REG_TYPE_NF, E(FLOAT) }, [BRW_REGISTER_TYPE_DF] = { GEN10_ALIGN1_3SRC_REG_TYPE_DF, E(FLOAT) }, [BRW_REGISTER_TYPE_F] = { GEN10_ALIGN1_3SRC_REG_TYPE_F, E(FLOAT) }, [BRW_REGISTER_TYPE_HF] = { GEN10_ALIGN1_3SRC_REG_TYPE_HF, E(FLOAT) }, @@ -306,6 +312,7 @@ unsigned brw_reg_type_to_size(enum brw_reg_type type) { static const unsigned type_size[] = { + [BRW_REGISTER_TYPE_NF] = 8, [BRW_REGISTER_TYPE_DF] = 8, [BRW_REGISTER_TYPE_F] = 4, [BRW_REGISTER_TYPE_HF] = 2, @@ -335,6 +342,7 @@ const char * brw_reg_type_to_letters(enum brw_reg_type type) { static const char letters[][3] = { + [BRW_REGISTER_TYPE_NF] = "NF", [BRW_REGISTER_TYPE_DF] = "DF", [BRW_REGISTER_TYPE_F] = "F", [BRW_REGISTER_TYPE_HF] = "HF", diff --git a/src/intel/compiler/brw_reg_type.h b/src/intel/compiler/brw_reg_type.h index 3bc9a36e7bf..ffbec90d3fe 100644 --- a/src/intel/compiler/brw_reg_type.h +++ b/src/intel/compiler/brw_reg_type.h @@ -45,6 +45,7 @@ struct gen_device_info; */ enum PACKED brw_reg_type { /** Floating-point types: @{ */ + BRW_REGISTER_TYPE_NF, BRW_REGISTER_TYPE_DF, BRW_REGISTER_TYPE_F, BRW_REGISTER_TYPE_HF, @@ -71,6 +72,7 @@ static inline bool brw_reg_type_is_floating_point(enum brw_reg_type type) { switch (type) { + case BRW_REGISTER_TYPE_NF: case BRW_REGISTER_TYPE_DF: case BRW_REGISTER_TYPE_F: case BRW_REGISTER_TYPE_HF: diff --git a/src/intel/compiler/brw_shader.cpp b/src/intel/compiler/brw_shader.cpp index 1df4f35cd8e..0e713819d94 100644 --- a/src/intel/compiler/brw_shader.cpp +++ b/src/intel/compiler/brw_shader.cpp @@ -541,6 +541,8 @@ brw_saturate_immediate(enum brw_reg_type type, struct brw_reg *reg) unreachable("unimplemented: saturate vector immediate"); case BRW_REGISTER_TYPE_HF: unreachable("unimplemented: saturate HF immediate"); + case BRW_REGISTER_TYPE_NF: + unreachable("no NF immediates"); } if (size < 8) { @@ -590,6 +592,8 @@ brw_negate_immediate(enum brw_reg_type type, struct brw_reg *reg) assert(!"unimplemented: negate UV/V immediate"); case BRW_REGISTER_TYPE_HF: assert(!"unimplemented: negate HF immediate"); + case BRW_REGISTER_TYPE_NF: + unreachable("no NF immediates"); } return false; @@ -632,6 +636,8 @@ brw_abs_immediate(enum brw_reg_type type, struct brw_reg *reg) assert(!"unimplemented: abs V immediate"); case BRW_REGISTER_TYPE_HF: assert(!"unimplemented: abs HF immediate"); + case BRW_REGISTER_TYPE_NF: + unreachable("no NF immediates"); } return false; -- 2.30.2