wine/vkd3d
0
0
Fork 0
mirror of https://gitlab.winehq.org/wine/vkd3d.git synced 2025-04-13 05:43:18 -07:00
Code Issues Packages Projects Releases Wiki Activity

vkd3d-shader/hlsl: Run folding passes again after lower_nonconstant_array_loads.

Browse Source 
This is because lower_nonconstant_array_loads() can potentially turn
nonconstant loads into constant loads, allowing copy-prop to turn these
loads into previous instructions, which might help other passes as well.

This patch lowers the number of required temps for the following ps_2_0
shader from 19 to 16:

    int i;
    float3x3 mats[4];

    float4 main() : sv_target
    {
        return mul(mats[i], float3(1, 2, 3)).xyzz;
    }
This commit is contained in:
Francisco Casas
2025-01-29 14:14:24 -03:00
committed by Henri Verbeet
parent e60c89c532
commit 153b7c8460
Notes: Henri Verbeet 2025-02-20 16:06:52 +01:00 
Approved-by: Henri Verbeet (@hverbeet)
Approved-by: Elizabeth Figura (@zfigura)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/1367
1 changed files with 21 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
libs/vkd3d-shader/hlsl_codegen.c
View File

@@ -7024,6 +7024,24 @@ void hlsl_lower_index_loads(struct hlsl_ctx *ctx, struct hlsl_block *body)
lower_ir(ctx, lower_index_loads, body); lower_ir(ctx, lower_index_loads, body);
} }
static void hlsl_run_folding_passes(struct hlsl_ctx *ctx, struct hlsl_block *body)
{
bool progress;
hlsl_transform_ir(ctx, fold_redundant_casts, body, NULL);
do
{
progress = hlsl_transform_ir(ctx, hlsl_fold_constant_exprs, body, NULL);
progress |= hlsl_transform_ir(ctx, hlsl_fold_constant_identities, body, NULL);
progress |= hlsl_transform_ir(ctx, hlsl_normalize_binary_exprs, body, NULL);
progress |= hlsl_transform_ir(ctx, hlsl_fold_constant_swizzles, body, NULL);
progress |= hlsl_copy_propagation_execute(ctx, body);
progress |= hlsl_transform_ir(ctx, fold_swizzle_chains, body, NULL);
progress |= hlsl_transform_ir(ctx, remove_trivial_swizzles, body, NULL);
progress |= hlsl_transform_ir(ctx, remove_trivial_conditional_branches, body, NULL);
} while (progress);
}
void hlsl_run_const_passes(struct hlsl_ctx *ctx, struct hlsl_block *body) void hlsl_run_const_passes(struct hlsl_ctx *ctx, struct hlsl_block *body)
{ {
bool progress; bool progress;
@@ -7048,19 +7066,8 @@ void hlsl_run_const_passes(struct hlsl_ctx *ctx, struct hlsl_block *body)
lower_ir(ctx, lower_int_abs, body); lower_ir(ctx, lower_int_abs, body);
lower_ir(ctx, lower_casts_to_bool, body); lower_ir(ctx, lower_casts_to_bool, body);
lower_ir(ctx, lower_float_modulus, body); lower_ir(ctx, lower_float_modulus, body);
hlsl_transform_ir(ctx, fold_redundant_casts, body, NULL);
do hlsl_run_folding_passes(ctx, body);
{
progress = hlsl_transform_ir(ctx, hlsl_fold_constant_exprs, body, NULL);
progress |= hlsl_transform_ir(ctx, hlsl_fold_constant_identities, body, NULL);
progress |= hlsl_transform_ir(ctx, hlsl_normalize_binary_exprs, body, NULL);
progress |= hlsl_transform_ir(ctx, hlsl_fold_constant_swizzles, body, NULL);
progress |= hlsl_copy_propagation_execute(ctx, body);
progress |= hlsl_transform_ir(ctx, fold_swizzle_chains, body, NULL);
progress |= hlsl_transform_ir(ctx, remove_trivial_swizzles, body, NULL);
progress |= hlsl_transform_ir(ctx, remove_trivial_conditional_branches, body, NULL);
} while (progress);
} }
static void generate_vsir_signature_entry(struct hlsl_ctx *ctx, struct vsir_program *program, static void generate_vsir_signature_entry(struct hlsl_ctx *ctx, struct vsir_program *program,
@@ -12577,6 +12584,8 @@ static void process_entry_function(struct hlsl_ctx *ctx,
lower_ir(ctx, validate_nonconstant_vector_store_derefs, body); lower_ir(ctx, validate_nonconstant_vector_store_derefs, body);
hlsl_run_folding_passes(ctx, body);
do do
compute_liveness(ctx, entry_func); compute_liveness(ctx, entry_func);
while (hlsl_transform_ir(ctx, dce, body, NULL)); while (hlsl_transform_ir(ctx, dce, body, NULL));