[PATCH 0/1] MR3694: d3dcompiler/tests: Avoid interpolation filters in test_samplers().

[Matteo Bruni (＠Mystral)]

To avoid accuracy issues, specifically with the QXL driver. In preparation of the next vkd3d merge when these tests will start to work in Wine.

[PATCH 1/1] d3dcompiler/tests: Avoid interpolation filters in test_samplers().

[Matteo Bruni]

From: Matteo Bruni

---
 dlls/d3dcompiler_43/tests/hlsl_d3d9.c | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/dlls/d3dcompiler_43/tests/hlsl_d3d9.c b/dlls/d3dcompiler_43/tests/hlsl_d3d9.c
index 42fb3b97fc2..c72aea3d024 100644
--- a/dlls/d3dcompiler_43/tests/hlsl_d3d9.c
+++ b/dlls/d3dcompiler_43/tests/hlsl_d3d9.c
@@ -1115,33 +1115,33 @@ static void test_samplers(void)
         "sampler s;\n"
         "float4 main() : COLOR\n"
         "{\n"
-        "    return tex2D(s, float2(0.5, 0.5));\n"
+        "    return tex2D(s, float2(0.75, 0.25));\n"
         "}",

"SamplerState s;\n"
         "float4 main() : COLOR\n"
         "{\n"
-        "    return tex2D(s, float2(0.5, 0.5));\n"
+        "    return tex2D(s, float2(0.75, 0.25));\n"
         "}",

"sampler2D s;\n"
         "float4 main() : COLOR\n"
         "{\n"
-        "    return tex2D(s, float2(0.5, 0.5));\n"
+        "    return tex2D(s, float2(0.75, 0.25));\n"
         "}",

"sampler s;\n"
         "Texture2D t;\n"
         "float4 main() : COLOR\n"
         "{\n"
-        "    return t.Sample(s, float2(0.5, 0.5));\n"
+        "    return t.Sample(s, float2(0.75, 0.25));\n"
         "}",

"SamplerState s;\n"
         "Texture2D t;\n"
         "float4 main() : COLOR\n"
         "{\n"
-        "    return t.Sample(s, float2(0.5, 0.5));\n"
+        "    return t.Sample(s, float2(0.75, 0.25));\n"
         "}",
     };

@@ -1161,7 +1161,7 @@ static void test_samplers(void)

hr = IDirect3DDevice9_SetTexture(test_context.device, 0, (IDirect3DBaseTexture9 *)texture);
     ok(hr == D3D_OK, "Got unexpected hr %#lx.\n", hr);
-    hr = IDirect3DDevice9_SetSamplerState(test_context.device, 0, D3DSAMP_MAGFILTER, D3DTEXF_LINEAR);
+    hr = IDirect3DDevice9_SetSamplerState(test_context.device, 0, D3DSAMP_MAGFILTER, D3DTEXF_POINT);
     ok(hr == D3D_OK, "Got unexpected hr %#lx.\n", hr);

for (i = 0; i < ARRAY_SIZE(tests); ++i)
@@ -1174,7 +1174,7 @@ static void test_samplers(void)
             draw_quad(test_context.device, ps_code);

v = get_color_vec4(test_context.device, 0, 0);
-            todo_wine ok(compare_vec4(&v, 0.25f, 0.0f, 0.25f, 0.0f, 128),
+            todo_wine ok(compare_vec4(&v, 1.0f, 0.0f, 1.0f, 0.0f, 0),
                     "Test %u: Got unexpected value {%.8e, %.8e, %.8e, %.8e}.\n", i, v.x, v.y, v.z, v.w);

ID3D10Blob_Release(ps_code);

[Zebediah Figura (＠zfigura)]

I guess, but we do test linear filtering in vkd3d, so it's not like this thoroughly avoids the issue. Just how inaccurate is QXL?

[Matteo Bruni (＠Mystral)]

On Thu Aug 31 17:51:27 2023 +0000, Zebediah Figura wrote:
> I guess, but we do test linear filtering in vkd3d, so it's not like this
> thoroughly avoids the issue. Just how inaccurate is QXL?

256K ULPs, which is crazy to me. See https://gitlab.winehq.org/wine/wine/-/merge_requests/3172#note_43940

Assuming there isn't some other explanation for the huge discrepancy, I think we should simply consider QXL not a suitable test platform. In this particular case, though, testing filtered sampling is not the point, while testing shader translation is, and the updated test does a better job of that as well IMO.

[PATCH 0/1] MR3694: d3dcompiler/tests: Avoid interpolation filters in test_samplers(). - approved

[Matteo Bruni (＠Mystral)]

This merge request was approved by Matteo Bruni.

[Zebediah Figura (＠zfigura)]

On Thu Aug 31 17:51:27 2023 +0000, Matteo Bruni wrote:
> 256K ULPs, which is crazy to me. See https://gitlab.winehq.org/wine/wine/-/merge_requests/3172#note_43940
> Assuming there isn't some other explanation for the huge discrepancy, I
> think we should simply consider QXL not a suitable test platform. In
> this particular case, though, testing filtered sampling is not the
> point, while testing shader translation is, and the updated test does a
> better job of that as well IMO.

Oh, ha ha, you know what it is? QXL is giving us 2.47058839e-001, and:

```
> > > 0x3f / 0xff

0.24705882352941178
```

It's rendering into an R8G8B8A8_UNORM render target and then converting that *back* to float. That can't possibly be in spec.

[Matteo Bruni (＠Mystral)]

On Thu Aug 31 18:29:16 2023 +0000, Zebediah Figura wrote:
> Oh, ha ha, you know what it is? QXL is giving us 2.47058839e-001, and:
> >>> 0x3f / 0xff
> 0.24705882352941178
>
> It's rendering into an R8G8B8A8_UNORM render target and then converting
> that *back* to float. That can't possibly be in spec.

That's 100% it :rolling_eyes:

But yeah, that's what I mean when I say that d3d QXL test results ought to be ignored.