radeonsi: Use TRUNC_COORD on samplers

The default behaviour (0) is: "round-nearest-even to n.6 and drop fraction when point sampling" whereas the OpenGL spec simply wants us to floor it (1) "truncate when point sampling".
See 8.14.2 in the OpenGL spec:

https://www.khronos.org/registry/OpenGL/specs/gl/glspec46.core.pdf

The Direct3D spec also mandates this (https://microsoft.github.io/DirectX-Specs/d3d/archive/D3D11_3_FunctionalSpec.htm#7.18.7%20Point%20Sample%20Addressing)

On WineD3D:
This fixes some point-sampling texture precision issues in some Direct3D 9 titles such as Guild Wars 2 and htoL#NiQ: The Firefly Diary that are not present on other vendors.

CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3953>

diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 08f32887216d95b38a97d02470f5f364252e11f2..19fb5a9cd1b2933eef72ddf1861817f78ac88b7b 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -4478,6 +4478,8 @@ static void *si_create_sampler_state(struct pipe_context *ctx,
    struct si_sampler_state *rstate = CALLOC_STRUCT(si_sampler_state);
    unsigned max_aniso = sscreen->force_aniso >= 0 ? sscreen->force_aniso : state->max_anisotropy;
    unsigned max_aniso_ratio = si_tex_aniso_filter(max_aniso);
+   bool trunc_coord = state->min_img_filter == PIPE_TEX_FILTER_NEAREST &&
+                      state->mag_img_filter == PIPE_TEX_FILTER_NEAREST;
    union pipe_color_union clamped_border_color;
 
    if (!rstate) {
@@ -4494,6 +4496,7 @@ static void *si_create_sampler_state(struct pipe_context *ctx,
        S_008F30_FORCE_UNNORMALIZED(!state->normalized_coords) |
        S_008F30_ANISO_THRESHOLD(max_aniso_ratio >> 1) | S_008F30_ANISO_BIAS(max_aniso_ratio) |
        S_008F30_DISABLE_CUBE_WRAP(!state->seamless_cube_map) |
+       S_008F30_TRUNC_COORD(trunc_coord) |
        S_008F30_COMPAT_MODE(sctx->chip_class == GFX8 || sctx->chip_class == GFX9));
    rstate->val[1] = (S_008F34_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 8)) |
                      S_008F34_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 8)) |