Nikolay Sivov : vkd3d-shader: Add constant folding for 'dp2add' operation.

[Alexandre Julliard]

Module: vkd3d
Branch: master
Commit: 7e99188dc700d258c8899b7560668326639cab0f
URL:    https://gitlab.winehq.org/wine/vkd3d/-/commit/7e99188dc700d258c8899b75606683...

Author: Nikolay Sivov 
Date:   Sun Jun 18 23:19:02 2023 +0200

vkd3d-shader: Add constant folding for 'dp2add' operation.

Signed-off-by: Nikolay Sivov

---

libs/vkd3d-shader/hlsl_constant_ops.c | 38 ++++++++++++++++++++++++++++++++++-
 1 file changed, 37 insertions(+), 1 deletion(-)

diff --git a/libs/vkd3d-shader/hlsl_constant_ops.c b/libs/vkd3d-shader/hlsl_constant_ops.c
index 5501b6e9..8b191de9 100644
--- a/libs/vkd3d-shader/hlsl_constant_ops.c
+++ b/libs/vkd3d-shader/hlsl_constant_ops.c
@@ -376,6 +376,36 @@ static bool fold_dot(struct hlsl_ctx *ctx, struct hlsl_constant_value *dst, cons
     return true;
 }

+static bool fold_dp2add(struct hlsl_ctx *ctx, struct hlsl_constant_value *dst, const struct hlsl_type *dst_type,
+        const struct hlsl_ir_constant *src1, const struct hlsl_ir_constant *src2, const struct hlsl_ir_constant *src3)
+{
+    enum hlsl_base_type type = dst_type->base_type;
+    unsigned int k;
+
+    assert(type == src1->node.data_type->base_type);
+    assert(type == src2->node.data_type->base_type);
+    assert(type == src3->node.data_type->base_type);
+    assert(src1->node.data_type->dimx == src2->node.data_type->dimx);
+    assert(src3->node.data_type->dimx == 1);
+
+    dst->u[0].f = src3->value.u[0].f;
+    for (k = 0; k < src1->node.data_type->dimx; ++k)
+    {
+        switch (type)
+        {
+            case HLSL_TYPE_FLOAT:
+            case HLSL_TYPE_HALF:
+                dst->u[0].f += src1->value.u[k].f * src2->value.u[k].f;
+                break;
+            default:
+                FIXME("Fold 'dp2add' for type %s.\n", debug_hlsl_type(ctx, dst_type));
+                return false;
+        }
+    }
+
+    return true;
+}
+
 static bool fold_div(struct hlsl_ctx *ctx, struct hlsl_constant_value *dst, const struct hlsl_type *dst_type,
         const struct hlsl_ir_constant *src1, const struct hlsl_ir_constant *src2,
         const struct vkd3d_shader_location *loc)
@@ -751,7 +781,7 @@ static bool fold_nequal(struct hlsl_ctx *ctx, struct hlsl_constant_value *dst, c

bool hlsl_fold_constant_exprs(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context)
 {
-    struct hlsl_ir_constant *arg1, *arg2 = NULL;
+    struct hlsl_ir_constant *arg1, *arg2 = NULL, *arg3 = NULL;
     struct hlsl_constant_value res = {0};
     struct hlsl_ir_node *res_node;
     struct hlsl_ir_expr *expr;
@@ -779,6 +809,8 @@ bool hlsl_fold_constant_exprs(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr,
     arg1 = hlsl_ir_constant(expr->operands[0].node);
     if (expr->operands[1].node)
         arg2 = hlsl_ir_constant(expr->operands[1].node);
+    if (expr->operands[2].node)
+        arg3 = hlsl_ir_constant(expr->operands[2].node);

switch (expr->op)
     {
@@ -856,6 +888,10 @@ bool hlsl_fold_constant_exprs(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr,
             success = fold_nequal(ctx, &res, instr->data_type, arg1, arg2);
             break;

+        case HLSL_OP3_DP2ADD:
+            success = fold_dp2add(ctx, &res, instr->data_type, arg1, arg2, arg3);
+            break;
+
         default:
             FIXME("Fold "%s" expression.\n", debug_hlsl_expr_op(expr->op));
             success = false;