October 2023 - Wine-commits

Alex Henrie : d3d8/tests: Use CRT allocation functions.
by Alexandre Julliard 31 Oct '23

31 Oct '23

Module: wine Branch: master Commit: 6db89702a3f7e7563c96f4ea3836b9cb261a8685 URL: https://gitlab.winehq.org/wine/wine/-/commit/6db89702a3f7e7563c96f4ea3836b9… Author: Alex Henrie <alexhenrie24(a)gmail.com> Date: Sun Oct 29 10:34:30 2023 -0600 d3d8/tests: Use CRT allocation functions. --- dlls/d3d8/tests/device.c | 43 +++++++++++++++++++++---------------------- dlls/d3d8/tests/stateblock.c | 40 ++++++++++++++++++++-------------------- dlls/d3d8/tests/visual.c | 8 ++++---- 3 files changed, 45 insertions(+), 46 deletions(-)

1 0

Nikolay Sivov : vkd3d-shader/hlsl: Add a pass to normalize switch cases blocks.
by Alexandre Julliard 31 Oct '23

31 Oct '23

Module: vkd3d Branch: master Commit: 68c14079a6bd59f1ec2e15e917b0df0592262cd8 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/68c14079a6bd59f1ec2e15e917b0d… Author: Nikolay Sivov <nsivov(a)codeweavers.com> Date: Wed Oct 11 14:39:01 2023 +0200 vkd3d-shader/hlsl: Add a pass to normalize switch cases blocks. Signed-off-by: Nikolay Sivov <nsivov(a)codeweavers.com> --- libs/vkd3d-shader/hlsl_codegen.c | 113 +++++++++++++++++++++++++++++++++++++++ tests/hlsl/switch.shader_test | 12 ++--- 2 files changed, 119 insertions(+), 6 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 013667e2..541a45ac 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -2169,6 +2169,118 @@ static bool remove_trivial_conditional_branches(struct hlsl_ctx *ctx, struct hls return true; } +static bool normalize_switch_cases(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) +{ + struct hlsl_ir_switch_case *c, *def = NULL; + bool missing_terminal_break = false; + struct hlsl_ir_node *node; + struct hlsl_ir_jump *jump; + struct hlsl_ir_switch *s; + + if (instr->type != HLSL_IR_SWITCH) + return false; + s = hlsl_ir_switch(instr); + + LIST_FOR_EACH_ENTRY(c, &s->cases, struct hlsl_ir_switch_case, entry) + { + bool terminal_break = false; + + if (list_empty(&c->body.instrs)) + { + terminal_break = !!list_next(&s->cases, &c->entry); + } + else + { + node = LIST_ENTRY(list_tail(&c->body.instrs), struct hlsl_ir_node, entry); + if (node->type == HLSL_IR_JUMP) + { + jump = hlsl_ir_jump(node); + terminal_break = jump->type == HLSL_IR_JUMP_BREAK; + } + } + + missing_terminal_break |= !terminal_break; + + if (!terminal_break) + { + if (c->is_default) + { + hlsl_error(ctx, &c->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_SYNTAX, + "The 'default' case block is not terminated with 'break' or 'return'."); + } + else + { + hlsl_error(ctx, &c->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_SYNTAX, + "Switch case block '%u' is not terminated with 'break' or 'return'.", c->value); + } + } + } + + if (missing_terminal_break) + return true; + + LIST_FOR_EACH_ENTRY(c, &s->cases, struct hlsl_ir_switch_case, entry) + { + if (c->is_default) + { + def = c; + + /* Remove preceding empty cases. */ + while (list_prev(&s->cases, &def->entry)) + { + c = LIST_ENTRY(list_prev(&s->cases, &def->entry), struct hlsl_ir_switch_case, entry); + if (!list_empty(&c->body.instrs)) + break; + hlsl_free_ir_switch_case(c); + } + + if (list_empty(&def->body.instrs)) + { + /* Remove following empty cases. */ + while (list_next(&s->cases, &def->entry)) + { + c = LIST_ENTRY(list_next(&s->cases, &def->entry), struct hlsl_ir_switch_case, entry); + if (!list_empty(&c->body.instrs)) + break; + hlsl_free_ir_switch_case(c); + } + + /* Merge with the next case. */ + if (list_next(&s->cases, &def->entry)) + { + c = LIST_ENTRY(list_next(&s->cases, &def->entry), struct hlsl_ir_switch_case, entry); + c->is_default = true; + hlsl_free_ir_switch_case(def); + def = c; + } + } + + break; + } + } + + if (def) + { + list_remove(&def->entry); + } + else + { + struct hlsl_ir_node *jump; + + if (!(def = hlsl_new_switch_case(ctx, 0, true, NULL, &s->node.loc))) + return true; + if (!(jump = hlsl_new_jump(ctx, HLSL_IR_JUMP_BREAK, NULL, &s->node.loc))) + { + hlsl_free_ir_switch_case(def); + return true; + } + hlsl_block_add_instr(&def->body, jump); + } + list_add_tail(&s->cases, &def->entry); + + return true; +} + static bool lower_nonconstant_vector_derefs(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, struct hlsl_block *block) { struct hlsl_ir_node *idx; @@ -4629,6 +4741,7 @@ int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry } while (progress); remove_unreachable_code(ctx, body); + hlsl_transform_ir(ctx, normalize_switch_cases, body, NULL); lower_ir(ctx, lower_nonconstant_vector_derefs, body); lower_ir(ctx, lower_casts_to_bool, body); diff --git a/tests/hlsl/switch.shader_test b/tests/hlsl/switch.shader_test index 720672a7..aab12485 100644 --- a/tests/hlsl/switch.shader_test +++ b/tests/hlsl/switch.shader_test @@ -211,7 +211,7 @@ float4 main() : sv_target } % unterminated cases -[pixel shader fail(sm<6) todo] +[pixel shader fail(sm<6)] uint4 v; float4 main() : sv_target @@ -230,7 +230,7 @@ float4 main() : sv_target return c; } -[pixel shader fail todo] +[pixel shader fail] uint4 v; float4 main() : sv_target @@ -246,7 +246,7 @@ float4 main() : sv_target return 0.0; } -[pixel shader fail todo] +[pixel shader fail] uint4 v; float4 main() : sv_target @@ -262,7 +262,7 @@ float4 main() : sv_target return 0.0; } -[pixel shader fail(sm<6) todo] +[pixel shader fail(sm<6)] uint4 v; float4 main() : sv_target @@ -279,7 +279,7 @@ float4 main() : sv_target return 0.0; } -[pixel shader fail(sm<6) todo] +[pixel shader fail(sm<6)] uint4 v; float4 main() : sv_target @@ -296,7 +296,7 @@ float4 main() : sv_target return 0.0; } -[pixel shader fail(sm<6) todo] +[pixel shader fail(sm<6)] uint4 v; float4 main() : sv_target

1 0

Nikolay Sivov : vkd3d-shader/tpf: Write out 'switch' statements.
by Alexandre Julliard 31 Oct '23

31 Oct '23

Module: vkd3d Branch: master Commit: bc2a4ee66a082d6892d7c01ec6a22a7e85a7ee91 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/bc2a4ee66a082d6892d7c01ec6a22… Author: Nikolay Sivov <nsivov(a)codeweavers.com> Date: Thu Oct 26 22:26:33 2023 +0200 vkd3d-shader/tpf: Write out 'switch' statements. Signed-off-by: Nikolay Sivov <nsivov(a)codeweavers.com> --- libs/vkd3d-shader/tpf.c | 44 +++++++++++++++++++++++++ tests/hlsl/switch.shader_test | 74 +++++++++++++++++++++---------------------- 2 files changed, 81 insertions(+), 37 deletions(-) diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index 0ea5a682..0be0ec4f 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -5467,6 +5467,46 @@ static void write_sm4_store(const struct tpf_writer *tpf, const struct hlsl_ir_s write_sm4_instruction(tpf, &instr); } +static void write_sm4_switch(const struct tpf_writer *tpf, const struct hlsl_ir_switch *s) +{ + const struct hlsl_ir_node *selector = s->selector.node; + struct hlsl_ir_switch_case *c; + struct sm4_instruction instr; + + memset(&instr, 0, sizeof(instr)); + instr.opcode = VKD3D_SM4_OP_SWITCH; + + sm4_src_from_node(tpf, &instr.srcs[0], selector, VKD3DSP_WRITEMASK_ALL); + instr.src_count = 1; + + write_sm4_instruction(tpf, &instr); + + LIST_FOR_EACH_ENTRY(c, &s->cases, struct hlsl_ir_switch_case, entry) + { + memset(&instr, 0, sizeof(instr)); + if (c->is_default) + { + instr.opcode = VKD3D_SM4_OP_DEFAULT; + } + else + { + struct hlsl_constant_value value = { .u[0].u = c->value }; + + instr.opcode = VKD3D_SM4_OP_CASE; + sm4_src_from_constant_value(&instr.srcs[0], &value, 1, VKD3DSP_WRITEMASK_ALL); + instr.src_count = 1; + } + + write_sm4_instruction(tpf, &instr); + write_sm4_block(tpf, &c->body); + } + + memset(&instr, 0, sizeof(instr)); + instr.opcode = VKD3D_SM4_OP_ENDSWITCH; + + write_sm4_instruction(tpf, &instr); +} + static void write_sm4_swizzle(const struct tpf_writer *tpf, const struct hlsl_ir_swizzle *swizzle) { unsigned int hlsl_swizzle; @@ -5554,6 +5594,10 @@ static void write_sm4_block(const struct tpf_writer *tpf, const struct hlsl_bloc write_sm4_store(tpf, hlsl_ir_store(instr)); break; + case HLSL_IR_SWITCH: + write_sm4_switch(tpf, hlsl_ir_switch(instr)); + break; + case HLSL_IR_SWIZZLE: write_sm4_swizzle(tpf, hlsl_ir_swizzle(instr)); break; diff --git a/tests/hlsl/switch.shader_test b/tests/hlsl/switch.shader_test index aab12485..243b0b11 100644 --- a/tests/hlsl/switch.shader_test +++ b/tests/hlsl/switch.shader_test @@ -1,7 +1,7 @@ [require] shader model >= 4.0 -[pixel shader todo] +[pixel shader] uint4 v; float4 main() : sv_target @@ -19,17 +19,17 @@ float4 main() : sv_target [test] uniform 0 uint4 3 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (5.0, 5.0, 5.0, 5.0) uniform 0 uint4 1 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (4.0, 4.0, 4.0, 4.0) uniform 0 uint4 0 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (3.0, 3.0, 3.0, 3.0) % falling through is only supported for empty case statements -[pixel shader todo] +[pixel shader] uint4 v; float4 main() : sv_target @@ -49,17 +49,17 @@ float4 main() : sv_target [test] uniform 0 uint4 2 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.0, 2.0, 3.0, 4.0) uniform 0 uint4 1 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.1, 2.0, 3.0, 4.0) uniform 0 uint4 0 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.1, 2.0, 3.0, 4.0) % case value evaluation -[pixel shader todo] +[pixel shader] uint4 v; float4 main() : sv_target @@ -81,14 +81,14 @@ float4 main() : sv_target [test] uniform 0 uint4 2 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.1, 2.1, 3.1, 4.1) uniform 0 uint4 1 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.0, 2.0, 3.0, 4.0) % floats are accepted -[pixel shader todo fail(sm>=6)] +[pixel shader fail(sm>=6)] uint4 v; float4 main() : sv_target @@ -110,13 +110,13 @@ float4 main() : sv_target [test] uniform 0 uint4 2 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.1, 2.1, 3.1, 4.1) uniform 0 uint4 1 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.0, 2.0, 3.0, 4.0) -[pixel shader todo fail(sm>=6)] +[pixel shader fail(sm>=6)] float4 v; float4 main() : sv_target @@ -138,10 +138,10 @@ float4 main() : sv_target [test] uniform 0 float4 2.0 0.0 0.0 0.0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.1, 2.1, 3.1, 4.1) uniform 0 float4 1.0 0.0 0.0 0.0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.0, 2.0, 3.0, 4.0) [pixel shader fail] @@ -312,7 +312,7 @@ float4 main() : sv_target } % more complicated breaks -[pixel shader todo] +[pixel shader] uint4 v; float4 main() : sv_target @@ -339,17 +339,17 @@ float4 main() : sv_target [test] uniform 0 uint4 2 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.1, 2.1, 3.1, 4.1) uniform 0 uint4 1 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.2, 2.2, 3.2, 4.2) uniform 0 uint4 0 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.0, 2.0, 3.0, 4.0) % switch breaks within a loop -[pixel shader todo] +[pixel shader] uint4 v; float4 main() : sv_target @@ -377,11 +377,11 @@ float4 main() : sv_target [test] uniform 0 uint4 2 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (5.0, 6.0, 7.0, 8.0) % default case placement -[pixel shader todo] +[pixel shader] uint4 v; float4 main() : sv_target @@ -408,16 +408,16 @@ float4 main() : sv_target [test] uniform 0 uint4 0 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (4.0, 5.0, 6.0, 7.0) uniform 0 uint4 2 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (2.0, 3.0, 4.0, 5.0) uniform 0 uint4 3 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (4.0, 5.0, 6.0, 7.0) -[pixel shader todo] +[pixel shader] uint4 v; float4 main() : sv_target @@ -445,13 +445,13 @@ float4 main() : sv_target [test] uniform 0 uint4 3 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.0, 2.0, 3.0, 4.0) uniform 0 uint4 0 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (4.0, 5.0, 6.0, 7.0) uniform 0 uint4 5 0 0 0 -todo draw quad +todo(sm>=6) draw quad probe all rgba (1.0, 2.0, 3.0, 4.0) % 'continue' is not supported in switches @@ -480,7 +480,7 @@ float4 main() : sv_target return c; } -[pixel shader todo] +[pixel shader] uint4 v; float4 main() : sv_target @@ -511,14 +511,14 @@ float4 main() : sv_target [test] uniform 0 uint4 0 0 3 1 -todo draw quad +todo(sm>=6) draw quad probe all rgba (10.0, 11.0, 12.0, 13.0) uniform 0 uint4 1 0 3 1 -todo draw quad +todo(sm>=6) draw quad probe all rgba (7.0, 8.0, 9.0, 10.0) % return from a switch nested in a loop -[pixel shader todo] +[pixel shader] uint4 v; float4 main() : sv_target @@ -545,8 +545,8 @@ float4 main() : sv_target [test] uniform 0 uint4 0 0 3 1 -todo draw quad +todo(sm>=6) draw quad probe all rgba (304.0, 305.0, 306.0, 307.0) uniform 0 uint4 1 0 3 1 -todo draw quad +todo(sm>=6) draw quad probe all rgba (3.0, 4.0, 5.0, 6.0)

1 0

Francisco Casas : tests: Test array indexing with multiple swizzles.
by Alexandre Julliard 31 Oct '23

31 Oct '23

Module: vkd3d Branch: master Commit: ab463f74bdfc286ccbb6a4adf0cdf2808d286f2a URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/ab463f74bdfc286ccbb6a4adf0cdf… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Mon Jul 10 10:41:37 2023 -0400 tests: Test array indexing with multiple swizzles. --- tests/hlsl/array-index-expr.shader_test | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) diff --git a/tests/hlsl/array-index-expr.shader_test b/tests/hlsl/array-index-expr.shader_test index 357b82f5..b7d91ea0 100644 --- a/tests/hlsl/array-index-expr.shader_test +++ b/tests/hlsl/array-index-expr.shader_test @@ -97,3 +97,22 @@ todo probe all rgba (5.0, 6.0, 7.0, 8.0) uniform 0 float4 1 1 0 0 todo draw quad todo probe all rgba (9.0, 10.0, 11.0, 12.0) + + +[pixel shader todo] +float4 a; + +float4 main() : sv_target +{ + float4 arr[] = {10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, 120}; + + float4 tmp = float4(1, 2, 3, 4); + tmp.yz = arr[a.z].wx; + + return tmp; +} + +[test] +uniform 0 float4 0 0 2.4 0 +todo draw quad +probe all rgba (1.0, 120.0, 90.0, 4.0)

1 0

Francisco Casas : vkd3d-shader/hlsl: Clean-up instruction block for offset node creation.
by Alexandre Julliard 31 Oct '23

31 Oct '23

Module: vkd3d Branch: master Commit: e93568f2900775a10c67d2358ef0fa53fcffead4 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/e93568f2900775a10c67d2358ef0f… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Wed Oct 11 18:13:31 2023 -0300 vkd3d-shader/hlsl: Clean-up instruction block for offset node creation. --- libs/vkd3d-shader/hlsl_codegen.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 541a45ac..1a477db8 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -29,8 +29,6 @@ static struct hlsl_ir_node *new_offset_from_path_index(struct hlsl_ctx *ctx, str struct hlsl_ir_node *idx_offset = NULL; struct hlsl_ir_node *c; - hlsl_block_init(block); - switch (type->class) { case HLSL_CLASS_VECTOR: @@ -111,9 +109,14 @@ static struct hlsl_ir_node *new_offset_instr_from_deref(struct hlsl_ctx *ctx, st { struct hlsl_block idx_block; + hlsl_block_init(&idx_block); + if (!(offset = new_offset_from_path_index(ctx, &idx_block, type, offset, deref->path[i].node, regset, loc))) + { + hlsl_block_cleanup(&idx_block); return NULL; + } hlsl_block_add_block(block, &idx_block);

1 0

Francisco Casas : vkd3d-shader/hlsl: Introduce hlsl_deref_is_lowered() helper.
by Alexandre Julliard 31 Oct '23

31 Oct '23

Module: vkd3d Branch: master Commit: 81be47c00bf1c92a18d4bb133af962623c666071 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/81be47c00bf1c92a18d4bb133af96… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Wed Oct 4 15:31:46 2023 -0300 vkd3d-shader/hlsl: Introduce hlsl_deref_is_lowered() helper. Some functions work with dereferences and need to know if they are lowered yet. This can be known checking if deref->offset.node is NULL or deref->data_type is NULL. I am using the latter since it keeps working even after the following patches that split deref->offset into constant and variable parts. --- libs/vkd3d-shader/hlsl.c | 20 +++++++------------- libs/vkd3d-shader/hlsl.h | 8 +++++++- libs/vkd3d-shader/hlsl_codegen.c | 4 +--- 3 files changed, 15 insertions(+), 17 deletions(-) diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index 29461e08..3e98fe39 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -249,14 +249,7 @@ static enum hlsl_regset type_get_regset(const struct hlsl_type *type) enum hlsl_regset hlsl_deref_get_regset(struct hlsl_ctx *ctx, const struct hlsl_deref *deref) { - struct hlsl_type *type; - - if (deref->data_type) - type = deref->data_type; - else - type = hlsl_deref_get_type(ctx, deref); - - return type_get_regset(type); + return type_get_regset(hlsl_deref_get_type(ctx, deref)); } unsigned int hlsl_type_get_sm4_offset(const struct hlsl_type *type, unsigned int offset) @@ -520,6 +513,7 @@ static bool init_deref(struct hlsl_ctx *ctx, struct hlsl_deref *deref, struct hl deref->var = var; deref->path_len = path_len; deref->offset.node = NULL; + deref->data_type = NULL; if (path_len == 0) { @@ -609,7 +603,7 @@ struct hlsl_type *hlsl_deref_get_type(struct hlsl_ctx *ctx, const struct hlsl_de assert(deref); - if (deref->offset.node) + if (hlsl_deref_is_lowered(deref)) return deref->data_type; type = deref->var->data_type; @@ -1120,7 +1114,7 @@ bool hlsl_copy_deref(struct hlsl_ctx *ctx, struct hlsl_deref *deref, const struc if (!other) return true; - assert(!other->offset.node); + assert(!hlsl_deref_is_lowered(other)); if (!init_deref(ctx, deref, other->var, other->path_len)) return false; @@ -1177,7 +1171,7 @@ struct hlsl_ir_node *hlsl_new_store_index(struct hlsl_ctx *ctx, const struct hls unsigned int i; assert(lhs); - assert(!lhs->offset.node); + assert(!hlsl_deref_is_lowered(lhs)); if (!(store = hlsl_alloc(ctx, sizeof(*store)))) return NULL; @@ -1384,7 +1378,7 @@ struct hlsl_ir_load *hlsl_new_load_index(struct hlsl_ctx *ctx, const struct hlsl struct hlsl_type *type; unsigned int i; - assert(!deref->offset.node); + assert(!hlsl_deref_is_lowered(deref)); type = hlsl_deref_get_type(ctx, deref); if (idx) @@ -1657,7 +1651,7 @@ static bool clone_deref(struct hlsl_ctx *ctx, struct clone_instr_map *map, { unsigned int i; - assert(!src->offset.node); + assert(!hlsl_deref_is_lowered(src)); if (!init_deref(ctx, dst, src->var, src->path_len)) return false; diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index bd09e869..328f5597 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -644,11 +644,17 @@ struct hlsl_deref * The path is lowered to this single offset -- whose value may vary between SM1 and SM4 -- * before writing the bytecode. * Since the type information cannot longer be retrieved from the offset alone, the type is - * stored in the data_type field. */ + * stored in the data_type field, which remains NULL if the deref hasn't been lowered yet. */ struct hlsl_src offset; struct hlsl_type *data_type; }; +/* Whether the path has been lowered to an offset or not. */ +static inline bool hlsl_deref_is_lowered(const struct hlsl_deref *deref) +{ + return !!deref->data_type; +} + struct hlsl_ir_load { struct hlsl_ir_node node; diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 1a477db8..71b88f1b 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -135,9 +135,7 @@ static bool replace_deref_path_with_offset(struct hlsl_ctx *ctx, struct hlsl_der struct hlsl_block block; assert(deref->var); - - /* register offsets shouldn't be used before this point is reached. */ - assert(!deref->offset.node); + assert(!hlsl_deref_is_lowered(deref)); type = hlsl_deref_get_type(ctx, deref);

1 0

Francisco Casas : vkd3d-shader/hlsl: Split deref-offset into a node and a constant uint.
by Alexandre Julliard 31 Oct '23

31 Oct '23

Module: vkd3d Branch: master Commit: 61a17643a2b27336f997a34642c853b6c3b44f52 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/61a17643a2b27336f997a34642c85… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Tue Oct 3 16:47:13 2023 -0300 vkd3d-shader/hlsl: Split deref-offset into a node and a constant uint. This uint will be used for the following: - Since SM4's relative addressing (the capability of passing a register as an index to another register) only has whole-register granularity, we will need to make the offset node express the offset in whole-registers and specify the register component in this uint, otherwise we would have to add additional / and % operations in the output binary. - If, after we apply constant folding and copy propagation, we determine that the offset is a single constant node, we can store all the offset in this uint constant, and remove the offset src. This allows DCE to remove a good bunch of the nodes previously required only for the offset constants, which makes the output more liteweight and readable, and simplifies the implementation of relative addressing when writing tpf in the following patches. In dump_deref(), we use "c" to indicate components instead of whole registers. Since now both the offset node and the offset uint are in components a lowered deref would look like: var[@42c + 2c] But, once we express the offset node in whole registers we will remove the "c" from the node part: var[@22 + 3c] --- libs/vkd3d-shader/hlsl.c | 33 ++++++++++++++----- libs/vkd3d-shader/hlsl.h | 12 ++++--- libs/vkd3d-shader/hlsl_codegen.c | 68 ++++++++++++++++++++++++---------------- 3 files changed, 74 insertions(+), 39 deletions(-)

1 0

Francisco Casas : vkd3d-shader/hlsl: Express deref->offset in whole registers.
by Alexandre Julliard 31 Oct '23

31 Oct '23

Module: vkd3d Branch: master Commit: 1520f327e5602041469816ff5992e5fe0c3e1224 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/1520f327e5602041469816ff5992e… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Tue Oct 3 19:14:54 2023 -0300 vkd3d-shader/hlsl: Express deref->offset in whole registers. This is required to use SM4 relative addressing, because it is limited to whole-register granularity. --- libs/vkd3d-shader/hlsl.c | 3 --- libs/vkd3d-shader/hlsl.h | 2 +- libs/vkd3d-shader/hlsl_codegen.c | 22 ++++++++++++---------- 3 files changed, 13 insertions(+), 14 deletions(-) diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index 437dd912..b26bde31 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -2437,10 +2437,7 @@ static void dump_deref(struct vkd3d_string_buffer *buffer, const struct hlsl_der vkd3d_string_buffer_printf(buffer, "["); if (show_rel) - { dump_src(buffer, &deref->offset); - vkd3d_string_buffer_printf(buffer, "c"); - } if (show_rel && show_const) vkd3d_string_buffer_printf(buffer, " + "); if (show_const) diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index 208615dd..39b57d2f 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -642,7 +642,7 @@ struct hlsl_deref * regset) from the start of the variable, to the part of the variable that is referenced. * This offset is stored using two fields, one for a variable part and other for a constant * part, which are added together: - * - offset: An offset given by an instruction node, in number of register components. + * - offset: An offset given by an instruction node, in whole registers. * - const_offset: A constant number of register components. * Since the type information cannot longer be retrieved from the offset alone, the type is * stored in the data_type field, which remains NULL if the deref hasn't been lowered yet. */ diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index f2389ad0..5c3b7eb6 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -37,14 +37,7 @@ static struct hlsl_ir_node *new_offset_from_path_index(struct hlsl_ctx *ctx, str case HLSL_CLASS_MATRIX: { - if (!(c = hlsl_new_uint_constant(ctx, 4, loc))) - return NULL; - hlsl_block_add_instr(block, c); - - if (!(idx_offset = hlsl_new_binary_expr(ctx, HLSL_OP2_MUL, c, idx))) - return NULL; - hlsl_block_add_instr(block, idx_offset); - + idx_offset = idx; break; } @@ -52,6 +45,12 @@ static struct hlsl_ir_node *new_offset_from_path_index(struct hlsl_ctx *ctx, str { unsigned int size = hlsl_type_get_array_element_reg_size(type->e.array.type, regset); + if (regset == HLSL_REGSET_NUMERIC) + { + assert(size % 4 == 0); + size /= 4; + } + if (!(c = hlsl_new_uint_constant(ctx, size, loc))) return NULL; hlsl_block_add_instr(block, c); @@ -73,7 +72,7 @@ static struct hlsl_ir_node *new_offset_from_path_index(struct hlsl_ctx *ctx, str { assert(*offset_component == 0); *offset_component = field_offset % 4; - field_offset -= *offset_component; + field_offset /= 4; } if (!(c = hlsl_new_uint_constant(ctx, field_offset, loc))) @@ -4470,7 +4469,10 @@ bool hlsl_offset_from_deref(struct hlsl_ctx *ctx, const struct hlsl_deref *deref if (offset_node->type != HLSL_IR_CONSTANT) return false; - *offset += hlsl_ir_constant(offset_node)->value.u[0].u; + if (regset == HLSL_REGSET_NUMERIC) + *offset += 4 * hlsl_ir_constant(offset_node)->value.u[0].u; + else + *offset += hlsl_ir_constant(offset_node)->value.u[0].u; } size = deref->var->data_type->reg_size[regset];

1 0

Francisco Casas : vkd3d-shader/hlsl: Absorb hlsl_ir_constant deref offsets into const_offset.
by Alexandre Julliard 31 Oct '23

31 Oct '23

Module: vkd3d Branch: master Commit: 74767beaf658296950f3f2ca527192e471d39bf4 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/74767beaf658296950f3f2ca52719… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Thu May 11 15:18:02 2023 -0400 vkd3d-shader/hlsl: Absorb hlsl_ir_constant deref offsets into const_offset. --- libs/vkd3d-shader/hlsl_codegen.c | 29 +++++++++++++++++++++-------- 1 file changed, 21 insertions(+), 8 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 5c3b7eb6..0a9d3f67 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -174,6 +174,24 @@ static bool replace_deref_path_with_offset(struct hlsl_ctx *ctx, struct hlsl_der return true; } +static bool clean_constant_deref_offset_srcs(struct hlsl_ctx *ctx, struct hlsl_deref *deref, + struct hlsl_ir_node *instr) +{ + if (deref->offset.node && deref->offset.node->type == HLSL_IR_CONSTANT) + { + enum hlsl_regset regset = hlsl_deref_get_regset(ctx, deref); + + if (regset == HLSL_REGSET_NUMERIC) + deref->const_offset += 4 * hlsl_ir_constant(deref->offset.node)->value.u[0].u; + else + deref->const_offset += hlsl_ir_constant(deref->offset.node)->value.u[0].u; + hlsl_src_remove(&deref->offset); + return true; + } + return false; +} + + /* Split uniforms into two variables representing the constant and temp * registers, and copy the former to the latter, so that writes to uniforms * work. */ @@ -4465,14 +4483,8 @@ bool hlsl_offset_from_deref(struct hlsl_ctx *ctx, const struct hlsl_deref *deref /* We should always have generated a cast to UINT. */ assert(offset_node->data_type->class == HLSL_CLASS_SCALAR && offset_node->data_type->base_type == HLSL_TYPE_UINT); - - if (offset_node->type != HLSL_IR_CONSTANT) - return false; - - if (regset == HLSL_REGSET_NUMERIC) - *offset += 4 * hlsl_ir_constant(offset_node)->value.u[0].u; - else - *offset += hlsl_ir_constant(offset_node)->value.u[0].u; + assert(offset_node->type != HLSL_IR_CONSTANT); + return false; } size = deref->var->data_type->reg_size[regset]; @@ -4789,6 +4801,7 @@ int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry /* TODO: move forward, remove when no longer needed */ transform_derefs(ctx, replace_deref_path_with_offset, body); while (hlsl_transform_ir(ctx, hlsl_fold_constant_exprs, body, NULL)); + transform_derefs(ctx, clean_constant_deref_offset_srcs, body); do compute_liveness(ctx, entry_func);

1 0

Francisco Casas : vkd3d-shader/hlsl: Rename hlsl_deref.offset to hlsl_deref.rel_offset.
by Alexandre Julliard 31 Oct '23

31 Oct '23

Module: vkd3d Branch: master Commit: 313df300ad317b49b45b38148755095988a974ed URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/313df300ad317b49b45b381487550… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Fri Oct 6 13:56:24 2023 -0300 vkd3d-shader/hlsl: Rename hlsl_deref.offset to hlsl_deref.rel_offset. This field is now analogous to vkd3d_shader_register_index.rel_addr. Also, it makes sense to rename it now because all the constant part of the offset is now handled to hlsl_deref.const_offset. Consequently, it may also be NULL now. --- libs/vkd3d-shader/hlsl.c | 12 ++++++------ libs/vkd3d-shader/hlsl.h | 4 ++-- libs/vkd3d-shader/hlsl_codegen.c | 36 ++++++++++++++++++------------------ 3 files changed, 26 insertions(+), 26 deletions(-) diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index b26bde31..f56ecc90 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -512,7 +512,7 @@ static bool init_deref(struct hlsl_ctx *ctx, struct hlsl_deref *deref, struct hl { deref->var = var; deref->path_len = path_len; - deref->offset.node = NULL; + deref->rel_offset.node = NULL; deref->const_offset = 0; deref->data_type = NULL; @@ -541,7 +541,7 @@ bool hlsl_init_deref_from_index_chain(struct hlsl_ctx *ctx, struct hlsl_deref *d deref->path = NULL; deref->path_len = 0; - deref->offset.node = NULL; + deref->rel_offset.node = NULL; deref->const_offset = 0; assert(chain); @@ -1138,7 +1138,7 @@ void hlsl_cleanup_deref(struct hlsl_deref *deref) deref->path = NULL; deref->path_len = 0; - hlsl_src_remove(&deref->offset); + hlsl_src_remove(&deref->rel_offset); deref->const_offset = 0; } @@ -2432,12 +2432,12 @@ static void dump_deref(struct vkd3d_string_buffer *buffer, const struct hlsl_der { bool show_rel, show_const; - show_rel = deref->offset.node; + show_rel = deref->rel_offset.node; show_const = deref->const_offset != 0 || !show_rel; vkd3d_string_buffer_printf(buffer, "["); if (show_rel) - dump_src(buffer, &deref->offset); + dump_src(buffer, &deref->rel_offset); if (show_rel && show_const) vkd3d_string_buffer_printf(buffer, " + "); if (show_const) @@ -3011,7 +3011,7 @@ static void free_ir_resource_load(struct hlsl_ir_resource_load *load) static void free_ir_resource_store(struct hlsl_ir_resource_store *store) { - hlsl_src_remove(&store->resource.offset); + hlsl_src_remove(&store->resource.rel_offset); hlsl_src_remove(&store->coords); hlsl_src_remove(&store->value); vkd3d_free(store); diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index 39b57d2f..38e605ba 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -642,11 +642,11 @@ struct hlsl_deref * regset) from the start of the variable, to the part of the variable that is referenced. * This offset is stored using two fields, one for a variable part and other for a constant * part, which are added together: - * - offset: An offset given by an instruction node, in whole registers. + * - rel_offset: An offset given by an instruction node, in whole registers. * - const_offset: A constant number of register components. * Since the type information cannot longer be retrieved from the offset alone, the type is * stored in the data_type field, which remains NULL if the deref hasn't been lowered yet. */ - struct hlsl_src offset; + struct hlsl_src rel_offset; unsigned int const_offset; struct hlsl_type *data_type; }; diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 0a9d3f67..4ea4f4cb 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -168,7 +168,7 @@ static bool replace_deref_path_with_offset(struct hlsl_ctx *ctx, struct hlsl_der list_move_before(&instr->entry, &block.instrs); hlsl_cleanup_deref(deref); - hlsl_src_from_node(&deref->offset, offset); + hlsl_src_from_node(&deref->rel_offset, offset); deref->const_offset = offset_component; return true; @@ -177,15 +177,15 @@ static bool replace_deref_path_with_offset(struct hlsl_ctx *ctx, struct hlsl_der static bool clean_constant_deref_offset_srcs(struct hlsl_ctx *ctx, struct hlsl_deref *deref, struct hlsl_ir_node *instr) { - if (deref->offset.node && deref->offset.node->type == HLSL_IR_CONSTANT) + if (deref->rel_offset.node && deref->rel_offset.node->type == HLSL_IR_CONSTANT) { enum hlsl_regset regset = hlsl_deref_get_regset(ctx, deref); if (regset == HLSL_REGSET_NUMERIC) - deref->const_offset += 4 * hlsl_ir_constant(deref->offset.node)->value.u[0].u; + deref->const_offset += 4 * hlsl_ir_constant(deref->rel_offset.node)->value.u[0].u; else - deref->const_offset += hlsl_ir_constant(deref->offset.node)->value.u[0].u; - hlsl_src_remove(&deref->offset); + deref->const_offset += hlsl_ir_constant(deref->rel_offset.node)->value.u[0].u; + hlsl_src_remove(&deref->rel_offset); return true; } return false; @@ -3298,8 +3298,8 @@ static void compute_liveness_recurse(struct hlsl_block *block, unsigned int loop if (!var->first_write) var->first_write = loop_first ? min(instr->index, loop_first) : instr->index; store->rhs.node->last_read = last_read; - if (store->lhs.offset.node) - store->lhs.offset.node->last_read = last_read; + if (store->lhs.rel_offset.node) + store->lhs.rel_offset.node->last_read = last_read; break; } case HLSL_IR_EXPR: @@ -3326,8 +3326,8 @@ static void compute_liveness_recurse(struct hlsl_block *block, unsigned int loop var = load->src.var; var->last_read = max(var->last_read, last_read); - if (load->src.offset.node) - load->src.offset.node->last_read = last_read; + if (load->src.rel_offset.node) + load->src.rel_offset.node->last_read = last_read; break; } case HLSL_IR_LOOP: @@ -3344,14 +3344,14 @@ static void compute_liveness_recurse(struct hlsl_block *block, unsigned int loop var = load->resource.var; var->last_read = max(var->last_read, last_read); - if (load->resource.offset.node) - load->resource.offset.node->last_read = last_read; + if (load->resource.rel_offset.node) + load->resource.rel_offset.node->last_read = last_read; if ((var = load->sampler.var)) { var->last_read = max(var->last_read, last_read); - if (load->sampler.offset.node) - load->sampler.offset.node->last_read = last_read; + if (load->sampler.rel_offset.node) + load->sampler.rel_offset.node->last_read = last_read; } if (load->coords.node) @@ -3376,8 +3376,8 @@ static void compute_liveness_recurse(struct hlsl_block *block, unsigned int loop var = store->resource.var; var->last_read = max(var->last_read, last_read); - if (store->resource.offset.node) - store->resource.offset.node->last_read = last_read; + if (store->resource.rel_offset.node) + store->resource.rel_offset.node->last_read = last_read; store->coords.node->last_read = last_read; store->value.node->last_read = last_read; break; @@ -4473,7 +4473,7 @@ bool hlsl_regset_index_from_deref(struct hlsl_ctx *ctx, const struct hlsl_deref bool hlsl_offset_from_deref(struct hlsl_ctx *ctx, const struct hlsl_deref *deref, unsigned int *offset) { enum hlsl_regset regset = hlsl_deref_get_regset(ctx, deref); - struct hlsl_ir_node *offset_node = deref->offset.node; + struct hlsl_ir_node *offset_node = deref->rel_offset.node; unsigned int size; *offset = deref->const_offset; @@ -4505,8 +4505,8 @@ unsigned int hlsl_offset_from_deref_safe(struct hlsl_ctx *ctx, const struct hlsl if (hlsl_offset_from_deref(ctx, deref, &offset)) return offset; - hlsl_fixme(ctx, &deref->offset.node->loc, "Dereference with non-constant offset of type %s.", - hlsl_node_type_to_string(deref->offset.node->type)); + hlsl_fixme(ctx, &deref->rel_offset.node->loc, "Dereference with non-constant offset of type %s.", + hlsl_node_type_to_string(deref->rel_offset.node->type)); return 0; }

1 0