[PATCH vkd3d v2 1/5] tests: Add some simple tests for HLSL function calls.

[Matteo Bruni]

From: Zebediah Figura

Signed-off-by: Zebediah Figura 
Signed-off-by: Giovanni Mascellani 
Signed-off-by: Francisco Casas 
Signed-off-by: Matteo Bruni 
---
v2: Get rid of the second comment entirely. I'm not sure that the C
standard really allows a void return, but regardless it's probably not
that important to bring that up here.

Makefile.am                     |   2 +
 tests/hlsl-function.shader_test | 164 ++++++++++++++++++++++++++++++++
 2 files changed, 166 insertions(+)
 create mode 100644 tests/hlsl-function.shader_test

diff --git a/Makefile.am b/Makefile.am
index 24257062..57cf5780 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -65,6 +65,7 @@ vkd3d_shader_tests = \
    tests/hlsl-cross.shader_test \
    tests/hlsl-duplicate-modifiers.shader_test \
    tests/hlsl-for.shader_test \
+	tests/hlsl-function.shader_test \
    tests/hlsl-function-overload.shader_test \
    tests/hlsl-gather-offset.shader_test \
    tests/hlsl-gather.shader_test \
@@ -316,6 +317,7 @@ XFAIL_TESTS = \
    tests/hlsl-bool-cast.shader_test \
    tests/hlsl-duplicate-modifiers.shader_test \
    tests/hlsl-for.shader_test \
+	tests/hlsl-function.shader_test \
    tests/hlsl-function-overload.shader_test \
    tests/hlsl-gather.shader_test \
    tests/hlsl-intrinsic-override.shader_test \
diff --git a/tests/hlsl-function.shader_test b/tests/hlsl-function.shader_test
new file mode 100644
index 00000000..8d17605b
--- /dev/null
+++ b/tests/hlsl-function.shader_test
@@ -0,0 +1,164 @@
+[pixel shader fail]
+
+float4 func();
+
+float4 main() : sv_target
+{
+    return func();
+}
+
+% It's legal to call an undefined function in unused code, though.
+
+[pixel shader]
+
+float4 func();
+
+float4 unused()
+{
+    return func();
+}
+
+float4 main() : sv_target
+{
+    return 0;
+}
+
+[pixel shader fail]
+
+void func(inout float o)
+{
+    o += 0.1;
+}
+
+float4 main() : sv_target
+{
+    float x = 0;
+    func(x + 1);
+    return 0;
+}
+
+[pixel shader fail]
+
+void func(inout float2 o)
+{
+    o += 0.1;
+}
+
+float4 main() : sv_target
+{
+    float2 x = 0;
+    func(x.yy);
+    return 0;
+}
+
+[pixel shader fail]
+
+void func(out float o)
+{
+    o = 0.1;
+}
+
+float4 main() : sv_target
+{
+    const float x = 0;
+    func(x);
+    return x;
+}
+
+[pixel shader fail]
+
+void func(inout float o)
+{
+}
+
+float4 main() : sv_target
+{
+    const float x = 0;
+    func(x);
+    return x;
+}
+
+[pixel shader fail]
+
+void func()
+{
+}
+
+float4 main() : sv_target
+{
+    return func();
+}
+
+[pixel shader fail]
+
+void foo()
+{
+}
+
+void bar()
+{
+    return foo();
+}
+
+float4 main() : sv_target
+{
+    bar();
+    return 0;
+}
+
+[pixel shader fail]
+
+float4 main() : sv_target
+{
+    func();
+    return 0;
+}
+
+void func()
+{
+}
+
+[pixel shader]
+
+float func(in float a, out float b, inout float c)
+{
+    c -= 0.2;
+    b = a * 2;
+    return a + 0.2;
+}
+
+float4 main() : sv_target
+{
+    float x[2], ret;
+
+    x[0] = 0.1;
+    x[1] = 0.9;
+    ret = func(0.3, x[0], x[1]);
+
+    return float4(ret, x[0], x[1], 0);
+}
+
+[test]
+draw quad
+probe all rgba (0.5, 0.6, 0.7, 0)
+
+[pixel shader]
+
+void func(in float a, inout float2 b)
+{
+    b.y += 0.1;
+    b *= a;
+}
+
+float4 main() : sv_target
+{
+    float3 q = float3(0.1, 0.2, 0.3);
+
+    func(3.0, q.zx);
+    func(0.5, q.yz);
+    return float4(q, 0);
+}
+
+[test]
+draw quad
+probe all rgba (0.6, 0.1, 0.5, 0)

[PATCH vkd3d v2 2/5] tests: Add some tests for casting output arguments to functions.

[Matteo Bruni]

From: Zebediah Figura

Signed-off-by: Zebediah Figura 
Signed-off-by: Giovanni Mascellani 
Signed-off-by: Matteo Bruni 
---
v2: Get rid of the test failing on SM4 only.

Makefile.am                          |  2 +
 tests/hlsl-function-cast.shader_test | 87 ++++++++++++++++++++++++++++
 2 files changed, 89 insertions(+)
 create mode 100644 tests/hlsl-function-cast.shader_test

diff --git a/Makefile.am b/Makefile.am
index 57cf5780..e4f92bb5 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -66,6 +66,7 @@ vkd3d_shader_tests = \
    tests/hlsl-duplicate-modifiers.shader_test \
    tests/hlsl-for.shader_test \
    tests/hlsl-function.shader_test \
+	tests/hlsl-function-cast.shader_test \
    tests/hlsl-function-overload.shader_test \
    tests/hlsl-gather-offset.shader_test \
    tests/hlsl-gather.shader_test \
@@ -318,6 +319,7 @@ XFAIL_TESTS = \
    tests/hlsl-duplicate-modifiers.shader_test \
    tests/hlsl-for.shader_test \
    tests/hlsl-function.shader_test \
+	tests/hlsl-function-cast.shader_test \
    tests/hlsl-function-overload.shader_test \
    tests/hlsl-gather.shader_test \
    tests/hlsl-intrinsic-override.shader_test \
diff --git a/tests/hlsl-function-cast.shader_test b/tests/hlsl-function-cast.shader_test
new file mode 100644
index 00000000..4f80ac59
--- /dev/null
+++ b/tests/hlsl-function-cast.shader_test
@@ -0,0 +1,87 @@
+% Test implicit and explicit casts on function output parameters.
+
+[pixel shader]
+
+uniform float4 f;
+
+void func(out float4 o)
+{
+    o = f;
+}
+
+float4 main() : sv_target
+{
+    int4 x;
+    func(x);
+    return x;
+}
+
+[test]
+uniform 0 float4 -1.9 -1.0 2.9 4.0
+draw quad
+probe all rgba (-1.0, -1.0, 2.0, 4.0)
+
+% As above, but cast "x" to float4 first.
+
+[pixel shader]
+
+uniform float4 f;
+
+void func(out float4 o)
+{
+    o = f;
+}
+
+float4 main() : sv_target
+{
+    int4 x;
+    func((float4)x);
+    return x;
+}
+
+[test]
+uniform 0 float4 -1.9 -1.0 2.9 4.0
+draw quad
+probe all rgba (-1.0, -1.0, 2.0, 4.0)
+
+% As above, but declare "x" as float4 and cast it to int4.
+
+[pixel shader]
+
+uniform float4 f;
+
+void func(out float4 o)
+{
+    o = f;
+}
+
+float4 main() : sv_target
+{
+    float4 x;
+    func((int4)x);
+    return x;
+}
+
+[test]
+uniform 0 float4 -1.9 -1.0 2.9 4.0
+draw quad
+probe all rgba (-1.0, -1.0, 2.0, 4.0)
+
+[pixel shader]
+
+void func(inout float4 a)
+{
+    a += 0.1;
+}
+
+float4 main(uniform int4 i) : sv_target
+{
+    int4 x = i;
+    func(x);
+    return x;
+}
+
+[test]
+uniform 0 int4 -2 0 1 -3000000
+draw quad
+probe all rgba (-1.0, 0.0, 1.0, -3000000.0)

[PATCH vkd3d v2 2/5] tests: Add some tests for casting output arguments to functions.

[Henri Verbeet]

Signed-off-by: Henri Verbeet

[PATCH vkd3d v2 3/5] vkd3d-shader/hlsl: Return false from hlsl_offset_from_deref() if the offset falls out of bounds.

[Matteo Bruni]

From: Zebediah Figura

Signed-off-by: Zebediah Figura 
Signed-off-by: Giovanni Mascellani 
Signed-off-by: Francisco Casas 
Signed-off-by: Matteo Bruni 
---
v2: Tweak the error message.

libs/vkd3d-shader/hlsl.h                 |  2 +-
 libs/vkd3d-shader/hlsl_codegen.c         | 25 ++++++++++++++++--------
 libs/vkd3d-shader/vkd3d_shader_private.h |  1 +
 3 files changed, 19 insertions(+), 9 deletions(-)

diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h
index be515e32..043d7fb3 100644
--- a/libs/vkd3d-shader/hlsl.h
+++ b/libs/vkd3d-shader/hlsl.h
@@ -790,7 +790,7 @@ unsigned int hlsl_combine_writemasks(unsigned int first, unsigned int second);
 unsigned int hlsl_map_swizzle(unsigned int swizzle, unsigned int writemask);
 unsigned int hlsl_swizzle_from_writemask(unsigned int writemask);

-bool hlsl_offset_from_deref(const struct hlsl_deref *deref, unsigned int *offset);
+bool hlsl_offset_from_deref(struct hlsl_ctx *ctx, const struct hlsl_deref *deref, unsigned int *offset);
 unsigned int hlsl_offset_from_deref_safe(struct hlsl_ctx *ctx, const struct hlsl_deref *deref);
 struct hlsl_reg hlsl_reg_from_deref(struct hlsl_ctx *ctx, const struct hlsl_deref *deref,
         const struct hlsl_type *type);
diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
index 17a7649f..e416647f 100644
--- a/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d-shader/hlsl_codegen.c
@@ -349,15 +349,16 @@ static void copy_propagation_set_value(struct copy_propagation_var_def *var_def,
     }
 }

-static struct hlsl_ir_node *copy_propagation_compute_replacement(const struct copy_propagation_state *state,
-        const struct hlsl_deref *deref, unsigned int count, unsigned int *swizzle)
+static struct hlsl_ir_node *copy_propagation_compute_replacement(struct hlsl_ctx *ctx,
+        const struct copy_propagation_state *state, const struct hlsl_deref *deref,
+        unsigned int count, unsigned int *swizzle)
 {
     const struct hlsl_ir_var *var = deref->var;
     struct copy_propagation_var_def *var_def;
     struct hlsl_ir_node *node = NULL;
     unsigned int offset, i;

-    if (!hlsl_offset_from_deref(deref, &offset))
+    if (!hlsl_offset_from_deref(ctx, deref, &offset))
         return NULL;

if (!(var_def = copy_propagation_get_var_def(state, var)))
@@ -414,7 +415,7 @@ static bool copy_propagation_transform_load(struct hlsl_ctx *ctx,
             return false;
     }

-    if (!(new_node = copy_propagation_compute_replacement(state, &load->src, dimx, &swizzle)))
+    if (!(new_node = copy_propagation_compute_replacement(ctx, state, &load->src, dimx, &swizzle)))
         return false;

if (type->type != HLSL_CLASS_OBJECT)
@@ -435,7 +436,7 @@ static bool copy_propagation_transform_object_load(struct hlsl_ctx *ctx,
     struct hlsl_ir_node *node;
     unsigned int swizzle;

-    if (!(node = copy_propagation_compute_replacement(state, deref, 1, &swizzle)))
+    if (!(node = copy_propagation_compute_replacement(ctx, state, deref, 1, &swizzle)))
         return false;

/* Only HLSL_IR_LOAD can produce an object. */
@@ -468,7 +469,7 @@ static void copy_propagation_record_store(struct hlsl_ctx *ctx, struct hlsl_ir_s
     if (!(var_def = copy_propagation_create_var_def(ctx, state, var)))
         return;

-    if (hlsl_offset_from_deref(lhs, &offset))
+    if (hlsl_offset_from_deref(ctx, lhs, &offset))
     {
         unsigned int writemask = store->writemask;

@@ -1546,7 +1547,7 @@ static bool type_is_single_reg(const struct hlsl_type *type)
     return type->type == HLSL_CLASS_SCALAR || type->type == HLSL_CLASS_VECTOR;
 }

-bool hlsl_offset_from_deref(const struct hlsl_deref *deref, unsigned int *offset)
+bool hlsl_offset_from_deref(struct hlsl_ctx *ctx, const struct hlsl_deref *deref, unsigned int *offset)
 {
     struct hlsl_ir_node *offset_node = deref->offset.node;

@@ -1564,6 +1565,14 @@ bool hlsl_offset_from_deref(const struct hlsl_deref *deref, unsigned int *offset
         return false;

*offset = hlsl_ir_constant(offset_node)->value[0].u;
+
+    if (*offset >= deref->var->data_type->reg_size)
+    {
+        hlsl_error(ctx, &deref->offset.node->loc, VKD3D_SHADER_ERROR_HLSL_OFFSET_OUT_OF_BOUNDS,
+                "Dereference is out of bounds.");
+        return false;
+    }
+
     return true;
 }

@@ -1571,7 +1580,7 @@ unsigned int hlsl_offset_from_deref_safe(struct hlsl_ctx *ctx, const struct hlsl
 {
     unsigned int offset;

-    if (hlsl_offset_from_deref(deref, &offset))
+    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.",
diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h
index 1de67f45..5ab8494c 100644
--- a/libs/vkd3d-shader/vkd3d_shader_private.h
+++ b/libs/vkd3d-shader/vkd3d_shader_private.h
@@ -115,6 +115,7 @@ enum vkd3d_shader_error
     VKD3D_SHADER_ERROR_HLSL_INVALID_RESERVATION         = 5016,
     VKD3D_SHADER_ERROR_HLSL_NOT_IMPLEMENTED             = 5017,
     VKD3D_SHADER_ERROR_HLSL_INVALID_TEXEL_OFFSET        = 5018,
+    VKD3D_SHADER_ERROR_HLSL_OFFSET_OUT_OF_BOUNDS        = 5019,

VKD3D_SHADER_WARNING_HLSL_IMPLICIT_TRUNCATION       = 5300,

[PATCH vkd3d v2 3/5] vkd3d-shader/hlsl: Return false from hlsl_offset_from_deref() if the offset falls out of bounds.

[Henri Verbeet]

Signed-off-by: Henri Verbeet

[PATCH vkd3d 4/5] vkd3d-shader/hlsl: Remove trivial swizzles after each copy prop pass.

[Matteo Bruni]

Signed-off-by: Matteo Bruni 
---
 Makefile.am                      | 1 -
 libs/vkd3d-shader/hlsl_codegen.c | 2 +-
 2 files changed, 1 insertion(+), 2 deletions(-)

diff --git a/Makefile.am b/Makefile.am
index e4f92bb5..f3db6d06 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -315,7 +315,6 @@ XFAIL_TESTS = \
    tests/hlsl-initializer-numeric.shader_test \
    tests/hlsl-initializer-static-array.shader_test \
    tests/hlsl-initializer-struct.shader_test \
-	tests/hlsl-bool-cast.shader_test \
    tests/hlsl-duplicate-modifiers.shader_test \
    tests/hlsl-for.shader_test \
    tests/hlsl-function.shader_test \
diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
index e416647f..2aa13427 100644
--- a/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d-shader/hlsl_codegen.c
@@ -1665,9 +1665,9 @@ int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_fun
     {
         progress = transform_ir(ctx, hlsl_fold_constants, body, NULL);
         progress |= copy_propagation_execute(ctx, body);
+        transform_ir(ctx, remove_trivial_swizzles, body, NULL);
     }
     while (progress);
-    transform_ir(ctx, remove_trivial_swizzles, body, NULL);

if (ctx->profile->major_version < 4)
         transform_ir(ctx, lower_division, body, NULL);

[PATCH vkd3d 4/5] vkd3d-shader/hlsl: Remove trivial swizzles after each copy prop pass.

[Giovanni Mascellani]

Hi,

Il 24/02/22 15:06, Matteo Bruni ha scritto:
> --- a/libs/vkd3d-shader/hlsl_codegen.c
> +++ b/libs/vkd3d-shader/hlsl_codegen.c
> @@ -1665,9 +1665,9 @@ int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_fun
>       {
>           progress = transform_ir(ctx, hlsl_fold_constants, body, NULL);
>           progress |= copy_propagation_execute(ctx, body);
> -
>    transform_ir(ctx, remove_trivial_swizzles, body, NULL);

Wouldn't it make sense to "progress |=" this one too? I guess that 
removing a trivial swizzle can enable another round of constant folding 
or copy propagation.

Giovanni.

[PATCH vkd3d 4/5] vkd3d-shader/hlsl: Remove trivial swizzles after each copy prop pass.

[Matteo Bruni]

On Thu, Feb 24, 2022 at 4:05 PM Giovanni Mascellani
 wrote:
> Hi,
>
> Il 24/02/22 15:06, Matteo Bruni ha scritto:
> > --- a/libs/vkd3d-shader/hlsl_codegen.c
> > +++ b/libs/vkd3d-shader/hlsl_codegen.c
> > @@ -1665,9 +1665,9 @@ int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_fun
> >       {
> >           progress = transform_ir(ctx, hlsl_fold_constants, body, NULL);
> >           progress |= copy_propagation_execute(ctx, body);
> > -
> >    transform_ir(ctx, remove_trivial_swizzles, body, NULL);
>
> Wouldn't it make sense to "progress |=" this one too? I guess that
> removing a trivial swizzle can enable another round of constant folding
> or copy propagation.

I remember thinking about it and dismissing it. Problem is, I don't
know how I reached that conclusion.

I'll resend this patch with the change.

[PATCH vkd3d v2 5/5] vkd3d-shader/hlsl: Correctly calculate offsets for array elements.

[Matteo Bruni]

From: Zebediah Figura

Signed-off-by: Matteo Bruni 
---
v2: Improve comment in the shader test.

Makefile.am                         |  2 +-
 libs/vkd3d-shader/hlsl.c            |  7 +++++++
 libs/vkd3d-shader/hlsl.h            |  1 +
 libs/vkd3d-shader/hlsl.y            |  2 +-
 libs/vkd3d-shader/hlsl_codegen.c    |  2 +-
 tests/hlsl-struct-array.shader_test | 19 +++++++++++++++++++
 6 files changed, 30 insertions(+), 3 deletions(-)
 create mode 100644 tests/hlsl-struct-array.shader_test

diff --git a/Makefile.am b/Makefile.am
index f3db6d06..85fadd38 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -93,6 +93,7 @@ vkd3d_shader_tests = \
    tests/hlsl-state-block-syntax.shader_test \
    tests/hlsl-static-initializer.shader_test \
    tests/hlsl-storage-qualifiers.shader_test \
+	tests/hlsl-struct-array.shader_test \
    tests/hlsl-struct-assignment.shader_test \
    tests/hlsl-struct-semantics.shader_test \
    tests/hlsl-vector-indexing.shader_test \
@@ -325,7 +326,6 @@ XFAIL_TESTS = \
    tests/hlsl-majority-pragma.shader_test \
    tests/hlsl-majority-typedef.shader_test \
    tests/hlsl-mul.shader_test \
-	tests/hlsl-nested-arrays.shader_test \
    tests/hlsl-numeric-constructor-truncation.shader_test \
    tests/hlsl-numeric-types.shader_test \
    tests/hlsl-operations.shader_test \
diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c
index a9a42770..a5aafdb9 100644
--- a/libs/vkd3d-shader/hlsl.c
+++ b/libs/vkd3d-shader/hlsl.c
@@ -202,6 +202,13 @@ static void hlsl_type_calculate_reg_size(struct hlsl_ctx *ctx, struct hlsl_type
     }
 }

+/* Returns the size of a type, considered as part of an array of that type.
+ * As such it includes padding after the type. */
+unsigned int hlsl_type_get_array_element_reg_size(const struct hlsl_type *type)
+{
+    return align(type->reg_size, 4);
+}
+
 static struct hlsl_type *hlsl_new_type(struct hlsl_ctx *ctx, const char *name, enum hlsl_type_class type_class,
         enum hlsl_base_type base_type, unsigned dimx, unsigned dimy)
 {
diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h
index 043d7fb3..8cce1273 100644
--- a/libs/vkd3d-shader/hlsl.h
+++ b/libs/vkd3d-shader/hlsl.h
@@ -782,6 +782,7 @@ bool hlsl_scope_add_type(struct hlsl_scope *scope, struct hlsl_type *type);
 struct hlsl_type *hlsl_type_clone(struct hlsl_ctx *ctx, struct hlsl_type *old,
         unsigned int default_majority, unsigned int modifiers);
 unsigned int hlsl_type_component_count(struct hlsl_type *type);
+unsigned int hlsl_type_get_array_element_reg_size(const struct hlsl_type *type);
 unsigned int hlsl_type_get_sm4_offset(const struct hlsl_type *type, unsigned int offset);
 bool hlsl_types_are_equal(const struct hlsl_type *t1, const struct hlsl_type *t2);

diff --git a/libs/vkd3d-shader/hlsl.y b/libs/vkd3d-shader/hlsl.y
index 459ad03b..70109780 100644
--- a/libs/vkd3d-shader/hlsl.y
+++ b/libs/vkd3d-shader/hlsl.y
@@ -607,7 +607,7 @@ static struct hlsl_ir_load *add_array_load(struct hlsl_ctx *ctx, struct list *in
         return NULL;
     }

-    if (!(c = hlsl_new_uint_constant(ctx, data_type->reg_size, loc)))
+    if (!(c = hlsl_new_uint_constant(ctx, hlsl_type_get_array_element_reg_size(data_type), loc)))
         return NULL;
     list_add_tail(instrs, &c->node.entry);
     if (!(mul = hlsl_new_binary_expr(ctx, HLSL_OP2_MUL, index, &c->node)))
diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
index 2aa13427..6a8c512f 100644
--- a/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d-shader/hlsl_codegen.c
@@ -621,7 +621,7 @@ static bool split_array_copies(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr,
     if (type->type != HLSL_CLASS_ARRAY)
         return false;
     element_type = type->e.array.type;
-    element_size = element_type->reg_size;
+    element_size = hlsl_type_get_array_element_reg_size(element_type);

for (i = 0; i < type->e.array.elements_count; ++i)
     {
diff --git a/tests/hlsl-struct-array.shader_test b/tests/hlsl-struct-array.shader_test
new file mode 100644
index 00000000..a5df9cdb
--- /dev/null
+++ b/tests/hlsl-struct-array.shader_test
@@ -0,0 +1,19 @@
+% In SM4, array elements are always aligned to the next vec4, although structs are not.
+
+[pixel shader]
+uniform struct
+{
+    float p, q;
+} colours[3];
+
+float4 main() : sv_target
+{
+    return float4(colours[0].q, colours[1].p, colours[2].q, colours[2].p);
+}
+
+[test]
+uniform 0 float4 0.1 0.2 0.0 0.0
+uniform 4 float4 0.3 0.4 0.0 0.0
+uniform 8 float4 0.5 0.6 0.0 0.0
+draw quad
+probe all rgba (0.2, 0.3, 0.6, 0.5)

[PATCH vkd3d v2 5/5] vkd3d-shader/hlsl: Correctly calculate offsets for array elements.

[Francisco Casas]

Signed-off-by: Francisco Casas

However, the previous patch

vkd3d-shader/hlsl: Remove trivial swizzles after each copy prop pass.

is required first, otherwise hlsl-nested-arrays.shader_test and 
hlsl-struct-array.shader_test fail.

Which I guess is why Matteo wrote it.

Best regards,
Francisco.

February 24, 2022 11:06 AM, "Matteo Bruni"  wrote:
> From: Zebediah Figura
> Signed-off-by: Matteo Bruni
>
> v2: Improve comment in the shader test.
>
> Makefile.am | 2 +-
> libs/vkd3d-shader/hlsl.c | 7 +++++++
> libs/vkd3d-shader/hlsl.h | 1 +
> libs/vkd3d-shader/hlsl.y | 2 +-
> libs/vkd3d-shader/hlsl_codegen.c | 2 +-
> tests/hlsl-struct-array.shader_test | 19 +++++++++++++++++++
> 6 files changed, 30 insertions(+), 3 deletions(-)
> create mode 100644 tests/hlsl-struct-array.shader_test
>
> diff --git a/Makefile.am b/Makefile.am
> index f3db6d06..85fadd38 100644
> --- a/Makefile.am
> +++ b/Makefile.am
> @@ -93,6 +93,7 @@ vkd3d_shader_tests = \
> tests/hlsl-state-block-syntax.shader_test \
> tests/hlsl-static-initializer.shader_test \
> tests/hlsl-storage-qualifiers.shader_test \
> - tests/hlsl-struct-array.shader_test \
>
> tests/hlsl-struct-assignment.shader_test \
> tests/hlsl-struct-semantics.shader_test \
> tests/hlsl-vector-indexing.shader_test \
> @@ -325,7 +326,6 @@ XFAIL_TESTS = \
> tests/hlsl-majority-pragma.shader_test \
> tests/hlsl-majority-typedef.shader_test \
> tests/hlsl-mul.shader_test \
> - tests/hlsl-nested-arrays.shader_test \
>
> tests/hlsl-numeric-constructor-truncation.shader_test \
> tests/hlsl-numeric-types.shader_test \
> tests/hlsl-operations.shader_test \
> diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c
> index a9a42770..a5aafdb9 100644
> --- a/libs/vkd3d-shader/hlsl.c
> +++ b/libs/vkd3d-shader/hlsl.c
> @@ -202,6 +202,13 @@ static void hlsl_type_calculate_reg_size(struct hlsl_ctx *ctx, struct
> hlsl_type
> }
> }
>
> +/* Returns the size of a type, considered as part of an array of that type.
> -
> - As such it includes padding after the type. */
>
> +unsigned int hlsl_type_get_array_element_reg_size(const struct hlsl_type *type)
> +{
> - return align(type->reg_size, 4);
>
> +}
> -
>
> static struct hlsl_type *hlsl_new_type(struct hlsl_ctx *ctx, const char *name, enum hlsl_type_class
> type_class,
> enum hlsl_base_type base_type, unsigned dimx, unsigned dimy)
> {
> diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h
> index 043d7fb3..8cce1273 100644
> --- a/libs/vkd3d-shader/hlsl.h
> +++ b/libs/vkd3d-shader/hlsl.h
> @@ -782,6 +782,7 @@ bool hlsl_scope_add_type(struct hlsl_scope *scope, struct hlsl_type *type);
> struct hlsl_type *hlsl_type_clone(struct hlsl_ctx *ctx, struct hlsl_type *old,
> unsigned int default_majority, unsigned int modifiers);
> unsigned int hlsl_type_component_count(struct hlsl_type *type);
> +unsigned int hlsl_type_get_array_element_reg_size(const struct hlsl_type *type);
> unsigned int hlsl_type_get_sm4_offset(const struct hlsl_type *type, unsigned int offset);
> bool hlsl_types_are_equal(const struct hlsl_type *t1, const struct hlsl_type *t2);
>
> diff --git a/libs/vkd3d-shader/hlsl.y b/libs/vkd3d-shader/hlsl.y
> index 459ad03b..70109780 100644
> --- a/libs/vkd3d-shader/hlsl.y
> +++ b/libs/vkd3d-shader/hlsl.y
> @@ -607,7 +607,7 @@ static struct hlsl_ir_load *add_array_load(struct hlsl_ctx *ctx, struct list
> *in
> return NULL;
> }
> - if (!(c = hlsl_new_uint_constant(ctx, data_type->reg_size, loc)))
> - if (!(c = hlsl_new_uint_constant(ctx, hlsl_type_get_array_element_reg_size(data_type), loc)))
>
> return NULL;
> list_add_tail(instrs, &c->node.entry);
> if (!(mul = hlsl_new_binary_expr(ctx, HLSL_OP2_MUL, index, &c->node)))
> diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
> index 2aa13427..6a8c512f 100644
> --- a/libs/vkd3d-shader/hlsl_codegen.c
> +++ b/libs/vkd3d-shader/hlsl_codegen.c
> @@ -621,7 +621,7 @@ static bool split_array_copies(struct hlsl_ctx *ctx, struct hlsl_ir_node
> *instr,
> if (type->type != HLSL_CLASS_ARRAY)
> return false;
> element_type = type->e.array.type;
> - element_size = element_type->reg_size;
> - element_size = hlsl_type_get_array_element_reg_size(element_type);
>
> for (i = 0; i < type->e.array.elements_count; ++i)
> {
> diff --git a/tests/hlsl-struct-array.shader_test b/tests/hlsl-struct-array.shader_test
> new file mode 100644
> index 00000000..a5df9cdb
> --- /dev/null
> +++ b/tests/hlsl-struct-array.shader_test
> @@ -0,0 +1,19 @@
> +% In SM4, array elements are always aligned to the next vec4, although structs are not.
> -
>
> +[pixel shader]
> +uniform struct
> +{
> - float p, q;
>
> +} colours[3];
> -
>
> +float4 main() : sv_target
> +{
> - return float4(colours[0].q, colours[1].p, colours[2].q, colours[2].p);
>
> +}
> -
>
> +[test]
> +uniform 0 float4 0.1 0.2 0.0 0.0
> +uniform 4 float4 0.3 0.4 0.0 0.0
> +uniform 8 float4 0.5 0.6 0.0 0.0
> +draw quad
> +probe all rgba (0.2, 0.3, 0.6, 0.5)
>
> 2.34.1

[Henri Verbeet]

Signed-off-by: Henri Verbeet