From: Francisco Casas fcasas@codeweavers.com
Signed-off-by: Francisco Casas fcasas@codeweavers.com Signed-off-by: Zebediah Figura zfigura@codeweavers.com --- v4: Move intrinsic_round() to be alphabetized with the other functions.
Makefile.am | 1 + libs/vkd3d-shader/hlsl.c | 1 + libs/vkd3d-shader/hlsl.h | 1 + libs/vkd3d-shader/hlsl.y | 8 ++++++++ libs/vkd3d-shader/hlsl_sm4.c | 4 ++++ tests/round.shader_test | 26 ++++++++++++++++++++++++++ 6 files changed, 41 insertions(+) create mode 100644 tests/round.shader_test
diff --git a/Makefile.am b/Makefile.am index 3186cdbf0..915174422 100644 --- a/Makefile.am +++ b/Makefile.am @@ -90,6 +90,7 @@ vkd3d_shader_tests = \ tests/preproc-invalid.shader_test \ tests/preproc-macro.shader_test \ tests/preproc-misc.shader_test \ + tests/round.shader_test \ tests/sampler.shader_test \ tests/saturate.shader_test \ tests/swizzle-0.shader_test \ diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index 1eee4278a..9824c56dc 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -1156,6 +1156,7 @@ const char *debug_hlsl_expr_op(enum hlsl_ir_expr_op op) [HLSL_OP1_NEG] = "-", [HLSL_OP1_NRM] = "nrm", [HLSL_OP1_RCP] = "rcp", + [HLSL_OP1_ROUND] = "round", [HLSL_OP1_RSQ] = "rsq", [HLSL_OP1_SAT] = "sat", [HLSL_OP1_SIGN] = "sign", diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index f2a7a48d5..04f86cdd6 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -299,6 +299,7 @@ enum hlsl_ir_expr_op HLSL_OP1_NEG, HLSL_OP1_NRM, HLSL_OP1_RCP, + HLSL_OP1_ROUND, HLSL_OP1_RSQ, HLSL_OP1_SAT, HLSL_OP1_SIGN, diff --git a/libs/vkd3d-shader/hlsl.y b/libs/vkd3d-shader/hlsl.y index b7e0409da..abf5da916 100644 --- a/libs/vkd3d-shader/hlsl.y +++ b/libs/vkd3d-shader/hlsl.y @@ -1611,6 +1611,12 @@ static bool intrinsic_pow(struct hlsl_ctx *ctx, return true; }
+static bool intrinsic_round(struct hlsl_ctx *ctx, + const struct parse_initializer *params, struct vkd3d_shader_location loc) +{ + return !!add_unary_arithmetic_expr(ctx, params->instrs, HLSL_OP1_ROUND, params->args[0], &loc); +} + static bool intrinsic_saturate(struct hlsl_ctx *ctx, const struct parse_initializer *params, struct vkd3d_shader_location loc) { @@ -1626,10 +1632,12 @@ static const struct intrinsic_function } intrinsic_functions[] = { + /* Note: these entries should be kept in alphabetical order. */ {"abs", 1, true, intrinsic_abs}, {"clamp", 3, true, intrinsic_clamp}, {"max", 2, true, intrinsic_max}, {"pow", 2, true, intrinsic_pow}, + {"round", 1, true, intrinsic_round}, {"saturate", 1, true, intrinsic_saturate}, };
diff --git a/libs/vkd3d-shader/hlsl_sm4.c b/libs/vkd3d-shader/hlsl_sm4.c index 413f33651..cd237efe5 100644 --- a/libs/vkd3d-shader/hlsl_sm4.c +++ b/libs/vkd3d-shader/hlsl_sm4.c @@ -1341,6 +1341,10 @@ static void write_sm4_expr(struct hlsl_ctx *ctx, write_sm4_unary_op(buffer, VKD3D_SM4_OP_MOV, &expr->node, arg1, VKD3D_SM4_REGISTER_MODIFIER_NEGATE); break;
+ case HLSL_OP1_ROUND: + write_sm4_unary_op(buffer, VKD3D_SM4_OP_ROUND_NE, &expr->node, arg1, 0); + break; + case HLSL_OP1_SAT: write_sm4_unary_op(buffer, VKD3D_SM4_OP_MOV | (VKD3D_SM4_INSTRUCTION_FLAG_SATURATE << VKD3D_SM4_INSTRUCTION_FLAGS_SHIFT), diff --git a/tests/round.shader_test b/tests/round.shader_test new file mode 100644 index 000000000..d8eb11b77 --- /dev/null +++ b/tests/round.shader_test @@ -0,0 +1,26 @@ +[pixel shader] +float4 main(uniform float4 u) : sv_target +{ + return round(u); +} + +[test] +uniform 0 float4 -0.5 6.5 7.5 3.4 +draw quad +probe all rgba (0.0, 6.0, 8.0, 3.0) 4 + + + +[pixel shader] +float4 main(uniform float4 u) : sv_target +{ + float a = round(u.r); + int2 b = round(u.gb); + float4 res = float4(b, a, u.a); + return round(res); +} + +[test] +uniform 0 float4 -0.5 6.5 7.5 3.4 +draw quad +probe all rgba (6.0, 8.0, 0.0, 3.0) 4
From: Francisco Casas fcasas@codeweavers.com
Signed-off-by: Francisco Casas fcasas@codeweavers.com Signed-off-by: Zebediah Figura zfigura@codeweavers.com --- v4: Rebase on top of the patch for round(), plus a couple of stylistic tweaks.
Makefile.am | 2 ++ libs/vkd3d-shader/hlsl.y | 55 ++++++++++++++++++++++++++++++++++++ tests/hlsl-cross.shader_test | 28 ++++++++++++++++++ 3 files changed, 85 insertions(+) create mode 100644 tests/hlsl-cross.shader_test
diff --git a/Makefile.am b/Makefile.am index 915174422..fb28d2103 100644 --- a/Makefile.am +++ b/Makefile.am @@ -64,6 +64,7 @@ vkd3d_shader_tests = \ tests/hlsl-bool-cast.shader_test \ tests/hlsl-clamp.shader_test \ tests/hlsl-comma.shader_test \ + tests/hlsl-cross.shader_test \ tests/hlsl-duplicate-modifiers.shader_test \ tests/hlsl-for.shader_test \ tests/hlsl-function-overload.shader_test \ @@ -288,6 +289,7 @@ XFAIL_TESTS = \ tests/conditional.shader_test \ tests/hlsl-array-dimension.shader_test \ tests/hlsl-bool-cast.shader_test \ + tests/hlsl-cross.shader_test \ tests/hlsl-duplicate-modifiers.shader_test \ tests/hlsl-for.shader_test \ tests/hlsl-function-overload.shader_test \ diff --git a/libs/vkd3d-shader/hlsl.y b/libs/vkd3d-shader/hlsl.y index abf5da916..ff35c09e5 100644 --- a/libs/vkd3d-shader/hlsl.y +++ b/libs/vkd3d-shader/hlsl.y @@ -1586,6 +1586,60 @@ static bool intrinsic_clamp(struct hlsl_ctx *ctx, return !!add_binary_arithmetic_expr(ctx, params->instrs, HLSL_OP2_MIN, &max->node, params->args[2], &loc); }
+static bool intrinsic_cross(struct hlsl_ctx *ctx, + const struct parse_initializer *params, struct vkd3d_shader_location loc) +{ + struct hlsl_ir_swizzle *arg1_swzl1, *arg1_swzl2, *arg2_swzl1, *arg2_swzl2; + struct hlsl_ir_node *arg1 = params->args[0], *arg2 = params->args[1]; + struct hlsl_ir_node *arg1_cast, *arg2_cast, *mul1_neg; + struct hlsl_ir_expr *mul1, *mul2; + struct hlsl_type *cast_type; + enum hlsl_base_type base; + + if (arg1->data_type->base_type == HLSL_TYPE_HALF && arg2->data_type->base_type == HLSL_TYPE_HALF) + base = HLSL_TYPE_HALF; + else + base = HLSL_TYPE_FLOAT; + + cast_type = hlsl_get_vector_type(ctx, base, 3); + + if (!(arg1_cast = add_implicit_conversion(ctx, params->instrs, arg1, cast_type, &loc))) + return false; + + if (!(arg2_cast = add_implicit_conversion(ctx, params->instrs, arg2, cast_type, &loc))) + return false; + + if (!(arg1_swzl1 = hlsl_new_swizzle(ctx, HLSL_SWIZZLE(Z, X, Y, Z), 3, arg1_cast, &loc))) + return false; + list_add_tail(params->instrs, &arg1_swzl1->node.entry); + + if (!(arg2_swzl1 = hlsl_new_swizzle(ctx, HLSL_SWIZZLE(Y, Z, X, Y), 3, arg2_cast, &loc))) + return false; + list_add_tail(params->instrs, &arg2_swzl1->node.entry); + + if (!(mul1 = add_binary_arithmetic_expr(ctx, params->instrs, HLSL_OP2_MUL, + &arg1_swzl1->node, &arg2_swzl1->node, &loc))) + return false; + + if (!(mul1_neg = hlsl_new_unary_expr(ctx, HLSL_OP1_NEG, &mul1->node, loc))) + return false; + list_add_tail(params->instrs, &mul1_neg->entry); + + if (!(arg1_swzl2 = hlsl_new_swizzle(ctx, HLSL_SWIZZLE(Y, Z, X, Y), 3, arg1_cast, &loc))) + return false; + list_add_tail(params->instrs, &arg1_swzl2->node.entry); + + if (!(arg2_swzl2 = hlsl_new_swizzle(ctx, HLSL_SWIZZLE(Z, X, Y, Z), 3, arg2_cast, &loc))) + return false; + list_add_tail(params->instrs, &arg2_swzl2->node.entry); + + if (!(mul2 = add_binary_arithmetic_expr(ctx, params->instrs, HLSL_OP2_MUL, + &arg1_swzl2->node, &arg2_swzl2->node, &loc))) + return false; + + return !!add_binary_arithmetic_expr(ctx, params->instrs, HLSL_OP2_ADD, &mul2->node, mul1_neg, &loc); +} + static bool intrinsic_max(struct hlsl_ctx *ctx, const struct parse_initializer *params, struct vkd3d_shader_location loc) { @@ -1635,6 +1689,7 @@ intrinsic_functions[] = /* Note: these entries should be kept in alphabetical order. */ {"abs", 1, true, intrinsic_abs}, {"clamp", 3, true, intrinsic_clamp}, + {"cross", 2, true, intrinsic_cross}, {"max", 2, true, intrinsic_max}, {"pow", 2, true, intrinsic_pow}, {"round", 1, true, intrinsic_round}, diff --git a/tests/hlsl-cross.shader_test b/tests/hlsl-cross.shader_test new file mode 100644 index 000000000..101db6071 --- /dev/null +++ b/tests/hlsl-cross.shader_test @@ -0,0 +1,28 @@ +[pixel shader] +float4 main(uniform float4 u, uniform float4 v) : sv_target +{ + float4 res = float4(0, 0, 0, 0); + res.xyz = cross(u, v); + return res; +} + +[test] +uniform 0 float4 1 -2 3 4 +uniform 4 float4 10 100 1000 10000 +draw quad +probe all rgba (-2300, -970, 120, 0) + + + +[pixel shader] +float4 main(uniform float4 u) : sv_target +{ + float4 res = float4(0, 0, 0, 3.5); + res.xyz = cross(u, 4); + return res; +} + +[test] +uniform 0 float4 1 -2 3 4 +draw quad +probe all rgba (-20, 8, 12, 3.5)
-
Zebediah Figura