diff --git a/libs/vkd3d-shader/hlsl.y b/libs/vkd3d-shader/hlsl.y
index 9cb14288..02a374f8 100644
--- a/libs/vkd3d-shader/hlsl.y
+++ b/libs/vkd3d-shader/hlsl.y
@@ -2128,8 +2128,10 @@ static bool add_assignment(struct hlsl_ctx *ctx, struct hlsl_block *block, struc
 
     if (hlsl_is_numeric_type(lhs_type))
     {
-        writemask = (1 << lhs_type->e.numeric.dimx) - 1;
-        width = lhs_type->e.numeric.dimx;
+        unsigned int size = hlsl_type_component_count(lhs_type);
+
+        writemask = (1 << size) - 1;
+        width = size;
     }
 
     if (!(rhs = add_implicit_conversion(ctx, block, rhs, lhs_type, &rhs->loc)))
@@ -2139,8 +2141,24 @@ static bool add_assignment(struct hlsl_ctx *ctx, struct hlsl_block *block, struc
     {
         if (lhs->type == HLSL_IR_EXPR && hlsl_ir_expr(lhs)->op == HLSL_OP1_CAST)
         {
-            hlsl_fixme(ctx, &lhs->loc, "Cast on the LHS.");
-            return false;
+            struct hlsl_ir_node *cast = lhs;
+            lhs = hlsl_ir_expr(cast)->operands[0].node;
+
+            if (hlsl_type_component_count(lhs->data_type) != hlsl_type_component_count(cast->data_type))
+            {
+                hlsl_fixme(ctx, &cast->loc, "Size change on the LHS.");
+                return false;
+            }
+            if (hlsl_version_ge(ctx, 4, 0))
+            {
+                hlsl_error(ctx, &cast->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_LVALUE,
+                        "Base type casts are not allowed on the LHS for profiles >= 4.");
+                return false;
+            }
+
+            lhs_type = lhs->data_type;
+            if (lhs_type->class == HLSL_CLASS_VECTOR || (lhs_type->class == HLSL_CLASS_MATRIX && matrix_writemask))
+                lhs_type = hlsl_get_vector_type(ctx, lhs->data_type->e.numeric.type, width);
         }
         else if (lhs->type == HLSL_IR_SWIZZLE)
         {
@@ -2181,6 +2199,7 @@ static bool add_assignment(struct hlsl_ctx *ctx, struct hlsl_block *block, struc
             hlsl_block_add_instr(block, new_swizzle);
 
             lhs = swizzle->val.node;
+            lhs_type = hlsl_get_vector_type(ctx, lhs_type->e.numeric.type, width);
             rhs = new_swizzle;
         }
         else
@@ -2190,6 +2209,12 @@ static bool add_assignment(struct hlsl_ctx *ctx, struct hlsl_block *block, struc
         }
     }
 
+    /* lhs casts could have resulted in a discrepancy between the
+     * rhs->data_type and the type of the variable that will be ulimately
+     * stored to. This is corrected. */
+    if (!(rhs = add_cast(ctx, block, rhs, lhs_type, &rhs->loc)))
+        return false;
+
     if (lhs->type == HLSL_IR_INDEX && hlsl_index_chain_has_resource_access(hlsl_ir_index(lhs)))
     {
         struct hlsl_ir_node *coords = hlsl_ir_index(lhs)->idx.node;
diff --git a/tests/hlsl/function-cast.shader_test b/tests/hlsl/function-cast.shader_test
index 4da6e0d9..a8135dc1 100644
--- a/tests/hlsl/function-cast.shader_test
+++ b/tests/hlsl/function-cast.shader_test
@@ -1,6 +1,6 @@
 % Test implicit and explicit casts on function output parameters.
 
-[pixel shader todo]
+[pixel shader todo(sm>=4)]
 
 uniform float4 f;
 
@@ -18,14 +18,14 @@ float4 main() : sv_target
 
 [test]
 uniform 0 float4 -1.9 -1.0 2.9 4.0
-todo(sm<6) draw quad
+todo(sm>=4 & sm<6) draw quad
 probe (0, 0) rgba (-1.0, -1.0, 2.0, 4.0)
 
 % As above, but cast "x" to float4 first.
 % In SM 6 a cast seems to implicitly promote the type to const,
 % so it fails to match the parameter of func().
 
-[pixel shader todo fail(sm>=6)]
+[pixel shader fail(sm>=6) todo(sm>=4)]
 
 uniform float4 f;
 
@@ -43,12 +43,12 @@ float4 main() : sv_target
 
 [test]
 uniform 0 float4 -1.9 -1.0 2.9 4.0
-todo draw quad
+todo(sm>=4) draw quad
 probe (0, 0) rgba (-1.0, -1.0, 2.0, 4.0)
 
 % As above, but declare "x" as float4 and cast it to int4.
 
-[pixel shader todo fail(sm>=6)]
+[pixel shader fail(sm>=6) todo(sm>=4)]
 
 uniform float4 f;
 
@@ -66,11 +66,11 @@ float4 main() : sv_target
 
 [test]
 uniform 0 float4 -1.9 -1.0 2.9 4.0
-todo draw quad
-probe (0, 0) rgba (-1.0, -1.0, 2.0, 4.0)
+todo(sm>=4) draw quad
+todo(sm<4) probe (0, 0) rgba (-1.0, -1.0, 2.0, 4.0)
 
 
-[pixel shader todo]
+[pixel shader todo(sm>=4)]
 uniform int4 i;
 
 void func(inout float4 a)
@@ -88,12 +88,12 @@ float4 main() : sv_target
 [test]
 if(sm<4)  uniform 0 float4 -2 0 1 -3000000
 if(sm>=4) uniform 0 int4 -2 0 1 -3000000
-todo(sm<6) draw quad
+todo(sm>=4 & sm<6) draw quad
 probe (0, 0) rgba (-1.0, 0.0, 1.0, -3000000.0) 4
 
 
 % An explicit cast gets applied right before assignment, as if it was on the lhs.
-[pixel shader todo fail(sm>=6)]
+[pixel shader fail(sm>=6) todo(sm>=4)]
 void fun(out float4 f)
 {
     f = float4(1.4, 2.6, 3.9, 4.3);
@@ -108,8 +108,8 @@ float4 main() : sv_target
 }
 
 [test]
-todo draw quad
-probe (0, 0) rgba(1, 2, 3, 4)
+todo(sm>=4) draw quad
+todo probe (0, 0) rgba(1, 2, 3, 4)
 
 
 [pixel shader todo fail]
@@ -176,7 +176,7 @@ float4 main() : sv_target
 }
 
 
-[pixel shader fail(sm>=4) todo]
+[pixel shader fail(sm>=4)]
 void fun(out float4 f)
 {
     f = float4(1.1, 2.3, 3.6, 4.3);
@@ -191,5 +191,5 @@ float4 main() : sv_target
 }
 
 [test]
-todo draw quad
+draw quad
 probe (0, 0) rgba(1.1, 2.3, 3.6, 4.3)
diff --git a/tests/hlsl/lhs-cast.shader_test b/tests/hlsl/lhs-cast.shader_test
index ff70bc08..7ced985d 100644
--- a/tests/hlsl/lhs-cast.shader_test
+++ b/tests/hlsl/lhs-cast.shader_test
@@ -14,7 +14,7 @@ probe (0, 0) rgba(1, 2, 3, 4)
 
 
 % Casts with base type changes are only valid on SM1.
-[pixel shader fail(sm>=4) todo]
+[pixel shader fail(sm>=4)]
 float4 main() : sv_target
 {
     float4 p = 0;
@@ -24,13 +24,13 @@ float4 main() : sv_target
 }
 
 [test]
-todo draw quad
+draw quad
 probe (0, 0) rgba(-1, -2, 4, 0)
 
 
 % Casts don't actually perform base type changes, only the outtermost one, which
 % I suspect is because the implicit cast on the assignment and not the cast itself.
-[pixel shader fail(sm>=4) todo]
+[pixel shader fail(sm>=4)]
 float4 main() : sv_target
 {
     float4 f = 0;
@@ -40,11 +40,11 @@ float4 main() : sv_target
 }
 
 [test]
-todo draw quad
+draw quad
 probe (0, 0) rgba(1.3, -2.4, 3.3, 4.7)
 
 
-[pixel shader fail(sm>=4) todo]
+[pixel shader fail(sm>=4)]
 float4 main() : sv_target
 {
     float4 f = 0;
@@ -54,7 +54,7 @@ float4 main() : sv_target
 }
 
 [test]
-todo draw quad
+draw quad
 probe (0, 0) rgba(1, -2, 3, 4)
 
 
@@ -69,7 +69,7 @@ float4 main() : sv_target
 }
 
 
-[pixel shader fail(sm>=6) todo]
+[pixel shader fail(sm>=6) todo(sm>=4)]
 float4 main() : sv_target
 {
     float4 f = 0;
@@ -79,11 +79,11 @@ float4 main() : sv_target
 }
 
 [test]
-todo draw quad
+todo(sm>=4) draw quad
 probe (0, 0) rgba(1, 2, 3, 4)
 
 
-[pixel shader fail(sm>=4) todo]
+[pixel shader fail(sm>=4)]
 float4 main() : sv_target
 {
     float4 f = 0;
@@ -93,11 +93,11 @@ float4 main() : sv_target
 }
 
 [test]
-todo draw quad
+draw quad
 probe (0, 0) rgba(1, 3, 0, 0);
 
 
-[pixel shader fail(sm>=4) todo]
+[pixel shader fail(sm>=4)]
 float4 main() : sv_target
 {
     float4 f = 0;
@@ -107,5 +107,5 @@ float4 main() : sv_target
 }
 
 [test]
-todo draw quad
+draw quad
 probe (0, 0) rgba(1.4, 2.8, 0, 0);
diff --git a/tests/hlsl/trigonometry.shader_test b/tests/hlsl/trigonometry.shader_test
index af21722d..b966a8aa 100644
--- a/tests/hlsl/trigonometry.shader_test
+++ b/tests/hlsl/trigonometry.shader_test
@@ -162,7 +162,7 @@ float4 main() : sv_target
     return float4(6789);
 }
 
-[pixel shader todo]
+[pixel shader todo(sm>=4)]
 float4 main() : sv_target
 {
     int sin_out, cos_out;
@@ -171,5 +171,5 @@ float4 main() : sv_target
 }
 
 [test]
-todo(sm<6) draw quad
+todo(sm>=4 & sm<6) draw quad
 probe (0, 0) rgba (0, 0, 0, 0);