From: Eric Anholt Date: Wed, 10 Jan 2018 20:48:05 +0000 (-0800) Subject: broadcom/vc5: Move V3D 3.3 texturing to a separate file. X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=42a35da96d69524dee262315bfcab3803f58bbff;p=mesa.git broadcom/vc5: Move V3D 3.3 texturing to a separate file. V3D 4.x texturing changes enough that #ifdefs would just make a mess of it. --- diff --git a/src/broadcom/Makefile.sources b/src/broadcom/Makefile.sources index 827c1944af6..e8f82fbeaac 100644 --- a/src/broadcom/Makefile.sources +++ b/src/broadcom/Makefile.sources @@ -29,6 +29,7 @@ BROADCOM_FILES = \ compiler/vir_to_qpu.c \ compiler/qpu_schedule.c \ compiler/qpu_validate.c \ + compiler/v3d33_tex.c \ compiler/v3d33_vpm_setup.c \ compiler/v3d_compiler.h \ compiler/v3d_nir_lower_io.c \ diff --git a/src/broadcom/compiler/meson.build b/src/broadcom/compiler/meson.build index 8a7211b649c..2c1b8d40b74 100644 --- a/src/broadcom/compiler/meson.build +++ b/src/broadcom/compiler/meson.build @@ -30,6 +30,7 @@ libbroadcom_compiler_files = files( 'vir_to_qpu.c', 'qpu_schedule.c', 'qpu_validate.c', + 'v3d33_tex.c', 'v3d33_vpm_setup.c', 'v3d_compiler.h', 'v3d_nir_lower_io.c', diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c index e76a87ca4cb..cec00cb55f7 100644 --- a/src/broadcom/compiler/nir_to_vir.c +++ b/src/broadcom/compiler/nir_to_vir.c @@ -32,15 +32,6 @@ #include "common/v3d_device_info.h" #include "v3d_compiler.h" -/* We don't do any address packing. */ -#define __gen_user_data void -#define __gen_address_type uint32_t -#define __gen_address_offset(reloc) (*reloc) -#define __gen_emit_reloc(cl, reloc) -#include "cle/v3d_packet_v33_pack.h" - -static struct qreg -ntq_get_src(struct v3d_compile *c, nir_src src, int i); static void ntq_emit_cf_list(struct v3d_compile *c, struct exec_list *list); @@ -65,7 +56,7 @@ resize_qreg_array(struct v3d_compile *c, (*regs)[i] = c->undef; } -static void +void vir_emit_thrsw(struct v3d_compile *c) { if (c->threads == 1) @@ -89,13 +80,6 @@ vir_SFU(struct v3d_compile *c, int waddr, struct qreg src) return vir_FMOV(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4)); } -static struct qreg -vir_LDTMU(struct v3d_compile *c) -{ - vir_NOP(c)->qpu.sig.ldtmu = true; - return vir_MOV(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4)); -} - static struct qreg indirect_uniform_load(struct v3d_compile *c, nir_intrinsic_instr *intr) { @@ -163,7 +147,7 @@ ntq_init_ssa_def(struct v3d_compile *c, nir_ssa_def *def) * (knowing that the previous instruction doesn't depend on flags) and rewrite * its destination to be the NIR reg's destination */ -static void +void ntq_store_dest(struct v3d_compile *c, nir_dest *dest, int chan, struct qreg result) { @@ -229,7 +213,7 @@ ntq_store_dest(struct v3d_compile *c, nir_dest *dest, int chan, } } -static struct qreg +struct qreg ntq_get_src(struct v3d_compile *c, nir_src src, int i) { struct hash_entry *entry; @@ -350,216 +334,7 @@ ntq_emit_tex(struct v3d_compile *c, nir_tex_instr *instr) break; } - struct V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1 p0_unpacked = { - V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1_header, - - .fetch_sample_mode = instr->op == nir_texop_txf, - }; - - struct V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1 p1_unpacked = { - }; - - switch (instr->sampler_dim) { - case GLSL_SAMPLER_DIM_1D: - if (instr->is_array) - p0_unpacked.lookup_type = TEXTURE_1D_ARRAY; - else - p0_unpacked.lookup_type = TEXTURE_1D; - break; - case GLSL_SAMPLER_DIM_2D: - case GLSL_SAMPLER_DIM_RECT: - if (instr->is_array) - p0_unpacked.lookup_type = TEXTURE_2D_ARRAY; - else - p0_unpacked.lookup_type = TEXTURE_2D; - break; - case GLSL_SAMPLER_DIM_3D: - p0_unpacked.lookup_type = TEXTURE_3D; - break; - case GLSL_SAMPLER_DIM_CUBE: - p0_unpacked.lookup_type = TEXTURE_CUBE_MAP; - break; - default: - unreachable("Bad sampler type"); - } - - struct qreg coords[5]; - int next_coord = 0; - for (unsigned i = 0; i < instr->num_srcs; i++) { - switch (instr->src[i].src_type) { - case nir_tex_src_coord: - for (int j = 0; j < instr->coord_components; j++) { - coords[next_coord++] = - ntq_get_src(c, instr->src[i].src, j); - } - if (instr->coord_components < 2) - coords[next_coord++] = vir_uniform_f(c, 0.5); - break; - case nir_tex_src_bias: - coords[next_coord++] = - ntq_get_src(c, instr->src[i].src, 0); - - p0_unpacked.bias_supplied = true; - break; - case nir_tex_src_lod: - coords[next_coord++] = - vir_FADD(c, - ntq_get_src(c, instr->src[i].src, 0), - vir_uniform(c, QUNIFORM_TEXTURE_FIRST_LEVEL, - unit)); - - if (instr->op != nir_texop_txf && - instr->op != nir_texop_tg4) { - p0_unpacked.disable_autolod_use_bias_only = true; - } - break; - case nir_tex_src_comparator: - coords[next_coord++] = - ntq_get_src(c, instr->src[i].src, 0); - - p0_unpacked.shadow = true; - break; - - case nir_tex_src_offset: { - nir_const_value *offset = - nir_src_as_const_value(instr->src[i].src); - p0_unpacked.texel_offset_for_s_coordinate = - offset->i32[0]; - - if (instr->coord_components >= 2) - p0_unpacked.texel_offset_for_t_coordinate = - offset->i32[1]; - - if (instr->coord_components >= 3) - p0_unpacked.texel_offset_for_r_coordinate = - offset->i32[2]; - break; - } - - default: - unreachable("unknown texture source"); - } - } - - bool return_16 = (c->key->tex[unit].return_size == 16 || - p0_unpacked.shadow); - - /* Limit the number of channels returned to both how many the NIR - * instruction writes and how many the instruction could produce. - */ - uint32_t instr_return_channels = nir_tex_instr_dest_size(instr); - if (return_16) - instr_return_channels = (instr_return_channels + 1) / 2; - - p1_unpacked.return_words_of_texture_data = - (1 << MIN2(instr_return_channels, - c->key->tex[unit].return_channels)) - 1; - - uint32_t p0_packed; - V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1_pack(NULL, - (uint8_t *)&p0_packed, - &p0_unpacked); - - uint32_t p1_packed; - V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1_pack(NULL, - (uint8_t *)&p1_packed, - &p1_unpacked); - /* Load unit number into the address field, which will be be used by - * the driver to decide which texture to put in the actual address - * field. - */ - p1_packed |= unit << 5; - - /* There is no native support for GL texture rectangle coordinates, so - * we have to rescale from ([0, width], [0, height]) to ([0, 1], [0, - * 1]). - */ - if (instr->sampler_dim == GLSL_SAMPLER_DIM_RECT) { - coords[0] = vir_FMUL(c, coords[0], - vir_uniform(c, QUNIFORM_TEXRECT_SCALE_X, - unit)); - coords[1] = vir_FMUL(c, coords[1], - vir_uniform(c, QUNIFORM_TEXRECT_SCALE_Y, - unit)); - } - - struct qreg texture_u[] = { - vir_uniform(c, QUNIFORM_TEXTURE_CONFIG_P0_0 + unit, p0_packed), - vir_uniform(c, QUNIFORM_TEXTURE_CONFIG_P1, p1_packed), - }; - uint32_t next_texture_u = 0; - - for (int i = 0; i < next_coord; i++) { - struct qreg dst; - - if (i == next_coord - 1) - dst = vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_TMUL); - else - dst = vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_TMU); - - struct qinst *tmu = vir_MOV_dest(c, dst, coords[i]); - - if (i < 2) { - tmu->has_implicit_uniform = true; - tmu->src[vir_get_implicit_uniform_src(tmu)] = - texture_u[next_texture_u++]; - } - } - - vir_emit_thrsw(c); - - struct qreg return_values[4]; - for (int i = 0; i < 4; i++) { - /* Swizzling .zw of an RG texture should give undefined - * results, not crash the compiler. - */ - if (p1_unpacked.return_words_of_texture_data & (1 << i)) - return_values[i] = vir_LDTMU(c); - else - return_values[i] = c->undef; - } - - for (int i = 0; i < nir_tex_instr_dest_size(instr); i++) { - struct qreg chan; - - if (return_16) { - STATIC_ASSERT(PIPE_SWIZZLE_X == 0); - chan = return_values[i / 2]; - - if (nir_alu_type_get_base_type(instr->dest_type) == - nir_type_float) { - enum v3d_qpu_input_unpack unpack; - if (i & 1) - unpack = V3D_QPU_UNPACK_H; - else - unpack = V3D_QPU_UNPACK_L; - - chan = vir_FMOV(c, chan); - vir_set_unpack(c->defs[chan.index], 0, unpack); - } else { - /* If we're unpacking the low field, shift it - * up to the top first. - */ - if ((i & 1) == 0) { - chan = vir_SHL(c, chan, - vir_uniform_ui(c, 16)); - } - - /* Do proper sign extension to a 32-bit int. */ - if (nir_alu_type_get_base_type(instr->dest_type) == - nir_type_int) { - chan = vir_ASR(c, chan, - vir_uniform_ui(c, 16)); - } else { - chan = vir_SHR(c, chan, - vir_uniform_ui(c, 16)); - } - } - } else { - chan = vir_MOV(c, return_values[i]); - } - ntq_store_dest(c, &instr->dest, i, chan); - } + v3d33_vir_emit_tex(c, instr); } static struct qreg diff --git a/src/broadcom/compiler/v3d33_tex.c b/src/broadcom/compiler/v3d33_tex.c new file mode 100644 index 00000000000..9af9285e07b --- /dev/null +++ b/src/broadcom/compiler/v3d33_tex.c @@ -0,0 +1,248 @@ +/* + * Copyright © 2016-2018 Broadcom + * + * 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. + */ + +#include "v3d_compiler.h" + +/* We don't do any address packing. */ +#define __gen_user_data void +#define __gen_address_type uint32_t +#define __gen_address_offset(reloc) (*reloc) +#define __gen_emit_reloc(cl, reloc) +#include "cle/v3d_packet_v33_pack.h" + +void +v3d33_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr) +{ + unsigned unit = instr->texture_index; + + struct V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1 p0_unpacked = { + V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1_header, + + .fetch_sample_mode = instr->op == nir_texop_txf, + }; + + struct V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1 p1_unpacked = { + }; + + switch (instr->sampler_dim) { + case GLSL_SAMPLER_DIM_1D: + if (instr->is_array) + p0_unpacked.lookup_type = TEXTURE_1D_ARRAY; + else + p0_unpacked.lookup_type = TEXTURE_1D; + break; + case GLSL_SAMPLER_DIM_2D: + case GLSL_SAMPLER_DIM_RECT: + if (instr->is_array) + p0_unpacked.lookup_type = TEXTURE_2D_ARRAY; + else + p0_unpacked.lookup_type = TEXTURE_2D; + break; + case GLSL_SAMPLER_DIM_3D: + p0_unpacked.lookup_type = TEXTURE_3D; + break; + case GLSL_SAMPLER_DIM_CUBE: + p0_unpacked.lookup_type = TEXTURE_CUBE_MAP; + break; + default: + unreachable("Bad sampler type"); + } + + struct qreg coords[5]; + int next_coord = 0; + for (unsigned i = 0; i < instr->num_srcs; i++) { + switch (instr->src[i].src_type) { + case nir_tex_src_coord: + for (int j = 0; j < instr->coord_components; j++) { + coords[next_coord++] = + ntq_get_src(c, instr->src[i].src, j); + } + if (instr->coord_components < 2) + coords[next_coord++] = vir_uniform_f(c, 0.5); + break; + case nir_tex_src_bias: + coords[next_coord++] = + ntq_get_src(c, instr->src[i].src, 0); + + p0_unpacked.bias_supplied = true; + break; + case nir_tex_src_lod: + coords[next_coord++] = + vir_FADD(c, + ntq_get_src(c, instr->src[i].src, 0), + vir_uniform(c, QUNIFORM_TEXTURE_FIRST_LEVEL, + unit)); + + if (instr->op != nir_texop_txf && + instr->op != nir_texop_tg4) { + p0_unpacked.disable_autolod_use_bias_only = true; + } + break; + case nir_tex_src_comparator: + coords[next_coord++] = + ntq_get_src(c, instr->src[i].src, 0); + + p0_unpacked.shadow = true; + break; + + case nir_tex_src_offset: { + nir_const_value *offset = + nir_src_as_const_value(instr->src[i].src); + p0_unpacked.texel_offset_for_s_coordinate = + offset->i32[0]; + + if (instr->coord_components >= 2) + p0_unpacked.texel_offset_for_t_coordinate = + offset->i32[1]; + + if (instr->coord_components >= 3) + p0_unpacked.texel_offset_for_r_coordinate = + offset->i32[2]; + break; + } + + default: + unreachable("unknown texture source"); + } + } + + bool return_16 = (c->key->tex[unit].return_size == 16 || + p0_unpacked.shadow); + + /* Limit the number of channels returned to both how many the NIR + * instruction writes and how many the instruction could produce. + */ + uint32_t instr_return_channels = nir_tex_instr_dest_size(instr); + if (return_16) + instr_return_channels = (instr_return_channels + 1) / 2; + + p1_unpacked.return_words_of_texture_data = + (1 << MIN2(instr_return_channels, + c->key->tex[unit].return_channels)) - 1; + + uint32_t p0_packed; + V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1_pack(NULL, + (uint8_t *)&p0_packed, + &p0_unpacked); + + uint32_t p1_packed; + V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1_pack(NULL, + (uint8_t *)&p1_packed, + &p1_unpacked); + /* Load unit number into the address field, which will be be used by + * the driver to decide which texture to put in the actual address + * field. + */ + p1_packed |= unit << 5; + + /* There is no native support for GL texture rectangle coordinates, so + * we have to rescale from ([0, width], [0, height]) to ([0, 1], [0, + * 1]). + */ + if (instr->sampler_dim == GLSL_SAMPLER_DIM_RECT) { + coords[0] = vir_FMUL(c, coords[0], + vir_uniform(c, QUNIFORM_TEXRECT_SCALE_X, + unit)); + coords[1] = vir_FMUL(c, coords[1], + vir_uniform(c, QUNIFORM_TEXRECT_SCALE_Y, + unit)); + } + + struct qreg texture_u[] = { + vir_uniform(c, QUNIFORM_TEXTURE_CONFIG_P0_0 + unit, p0_packed), + vir_uniform(c, QUNIFORM_TEXTURE_CONFIG_P1, p1_packed), + }; + uint32_t next_texture_u = 0; + + for (int i = 0; i < next_coord; i++) { + struct qreg dst; + + if (i == next_coord - 1) + dst = vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_TMUL); + else + dst = vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_TMU); + + struct qinst *tmu = vir_MOV_dest(c, dst, coords[i]); + + if (i < 2) { + tmu->has_implicit_uniform = true; + tmu->src[vir_get_implicit_uniform_src(tmu)] = + texture_u[next_texture_u++]; + } + } + + vir_emit_thrsw(c); + + struct qreg return_values[4]; + for (int i = 0; i < 4; i++) { + /* Swizzling .zw of an RG texture should give undefined + * results, not crash the compiler. + */ + if (p1_unpacked.return_words_of_texture_data & (1 << i)) + return_values[i] = vir_LDTMU(c); + else + return_values[i] = c->undef; + } + + for (int i = 0; i < nir_tex_instr_dest_size(instr); i++) { + struct qreg chan; + + if (return_16) { + STATIC_ASSERT(PIPE_SWIZZLE_X == 0); + chan = return_values[i / 2]; + + if (nir_alu_type_get_base_type(instr->dest_type) == + nir_type_float) { + enum v3d_qpu_input_unpack unpack; + if (i & 1) + unpack = V3D_QPU_UNPACK_H; + else + unpack = V3D_QPU_UNPACK_L; + + chan = vir_FMOV(c, chan); + vir_set_unpack(c->defs[chan.index], 0, unpack); + } else { + /* If we're unpacking the low field, shift it + * up to the top first. + */ + if ((i & 1) == 0) { + chan = vir_SHL(c, chan, + vir_uniform_ui(c, 16)); + } + + /* Do proper sign extension to a 32-bit int. */ + if (nir_alu_type_get_base_type(instr->dest_type) == + nir_type_int) { + chan = vir_ASR(c, chan, + vir_uniform_ui(c, 16)); + } else { + chan = vir_SHR(c, chan, + vir_uniform_ui(c, 16)); + } + } + } else { + chan = vir_MOV(c, return_values[i]); + } + ntq_store_dest(c, &instr->dest, i, chan); + } +} diff --git a/src/broadcom/compiler/v3d_compiler.h b/src/broadcom/compiler/v3d_compiler.h index 15cc16ceb60..cb501695aed 100644 --- a/src/broadcom/compiler/v3d_compiler.h +++ b/src/broadcom/compiler/v3d_compiler.h @@ -665,6 +665,10 @@ bool vir_writes_r3(const struct v3d_device_info *devinfo, struct qinst *inst); bool vir_writes_r4(const struct v3d_device_info *devinfo, struct qinst *inst); struct qreg vir_follow_movs(struct v3d_compile *c, struct qreg reg); uint8_t vir_channels_written(struct qinst *inst); +struct qreg ntq_get_src(struct v3d_compile *c, nir_src src, int i); +void ntq_store_dest(struct v3d_compile *c, nir_dest *dest, int chan, + struct qreg result); +void vir_emit_thrsw(struct v3d_compile *c); void vir_dump(struct v3d_compile *c); void vir_dump_inst(struct v3d_compile *c, struct qinst *inst); @@ -686,6 +690,7 @@ void vir_lower_uniforms(struct v3d_compile *c); void v3d33_vir_vpm_read_setup(struct v3d_compile *c, int num_components); void v3d33_vir_vpm_write_setup(struct v3d_compile *c); +void v3d33_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr); void v3d_vir_to_qpu(struct v3d_compile *c, struct qpu_reg *temp_registers); uint32_t v3d_qpu_schedule_instructions(struct v3d_compile *c); @@ -887,6 +892,14 @@ vir_NOP(struct v3d_compile *c) return vir_emit_nondef(c, vir_add_inst(V3D_QPU_A_NOP, c->undef, c->undef, c->undef)); } + +static inline struct qreg +vir_LDTMU(struct v3d_compile *c) +{ + vir_NOP(c)->qpu.sig.ldtmu = true; + return vir_MOV(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4)); +} + /* static inline struct qreg vir_LOAD_IMM(struct v3d_compile *c, uint32_t val)