tests/hlsl: Add float64 wave op tests.

Approved-by: Giovanni Mascellani (@giomasce)
Approved-by: Henri Verbeet (@hverbeet)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/820

tests/hlsl/wave-ops-float.shader_test

@@ -461,3 +461,239 @@ probe rtv 0 (1, 1) rgba (1.0, 0.0, 0.0, 1.0)
 probe uav 1 (1) rgba (0.5, 0.25, 1.0, 0.75)
 probe uav 1 (2) rgba (0.5, 0.25, 1.0, 0.75)
 probe uav 1 (3) rgba (0.5, 0.25, 1.0, 0.75)
+
+
+[require]
+shader model >= 6.0
+wave-ops
+float64
+format r32g32-float uav-load
+format r32g32b32a32-uint uav-load
+
+[uav 0]
+format r32g32-float
+size (buffer, 4)
+
+3.5 1.0
+3.5 1.0
+3.5 1.5
+3.5 1.0
+
+[uav 1]
+format r32g32b32a32-uint
+size (buffer, 16)
+
+0 0 0 0 0 0 0 0
+0 0 0 0 0 0 0 0
+0 0 0 0 0 0 0 0
+0 0 0 0 0 0 0 0
+0 0 0 0 0 0 0 0
+0 0 0 0 0 0 0 0
+0 0 0 0 0 0 0 0
+0 0 0 0 0 0 0 0
+
+[compute shader]
+RWBuffer<float2> u0;
+RWBuffer<uint4> u1;
+
+    [numthreads(4, 1, 1)]
+void main(uint id : SV_GroupIndex)
+{
+    double2 d = u0[id];
+    bool2 b = WaveActiveAllEqual(d);
+    u1[id] = uint4(b.x, b.y, 0, 0);
+}
+
+[test]
+dispatch 4 1 1
+probe uav 1 (0) rgbaui(1, 0, 0, 0)
+probe uav 1 (1) rgbaui(1, 0, 0, 0)
+probe uav 1 (2) rgbaui(1, 0, 0, 0)
+probe uav 1 (3) rgbaui(1, 0, 0, 0)
+
+
+[compute shader]
+RWBuffer<float2> u0;
+RWBuffer<double2> u1;
+
+    [numthreads(4, 1, 1)]
+void main(uint id : SV_GroupIndex)
+{
+    double2 d = u0[id] + u0[id ^ 1];
+    u1[id] = WaveReadLaneFirst(d);
+    u1[4 + id] = WaveReadLaneAt(d, 3);
+}
+
+[test]
+dispatch 4 1 1
+probe uav 1 (0) rgd(7.0, 2.0)
+probe uav 1 (1) rgd(7.0, 2.0)
+probe uav 1 (2) rgd(7.0, 2.0)
+probe uav 1 (3) rgd(7.0, 2.0)
+probe uav 1 (4) rgd(7.0, 2.5)
+probe uav 1 (5) rgd(7.0, 2.5)
+probe uav 1 (6) rgd(7.0, 2.5)
+probe uav 1 (7) rgd(7.0, 2.5)
+
+
+[compute shader]
+RWBuffer<float2> u0;
+RWBuffer<double2> u1;
+
+    [numthreads(4, 1, 1)]
+void main(uint id : SV_GroupIndex)
+{
+    double2 d = u0[id];
+    u1[id] = WaveActiveSum(d);
+}
+
+[test]
+dispatch 4 1 1
+probe uav 1 (0) rgd(14.0, 4.5)
+probe uav 1 (1) rgd(14.0, 4.5)
+probe uav 1 (2) rgd(14.0, 4.5)
+probe uav 1 (3) rgd(14.0, 4.5)
+
+
+[compute shader]
+RWBuffer<float2> u0;
+RWBuffer<double2> u1;
+
+    [numthreads(4, 1, 1)]
+void main(uint id : SV_GroupIndex)
+{
+    double2 d = u0[id];
+    u1[id] = WaveActiveProduct(d);
+}
+
+[test]
+dispatch 4 1 1
+probe uav 1 (0) rgd(150.0625, 1.5)
+probe uav 1 (1) rgd(150.0625, 1.5)
+probe uav 1 (2) rgd(150.0625, 1.5)
+probe uav 1 (3) rgd(150.0625, 1.5)
+
+
+[compute shader]
+RWBuffer<float2> u0;
+RWBuffer<double2> u1;
+
+    [numthreads(4, 1, 1)]
+void main(uint id : SV_GroupIndex)
+{
+    double2 d = u0[id];
+    u1[id] = WaveActiveMin(d);
+}
+
+[test]
+dispatch 4 1 1
+probe uav 1 (0) rgd(3.5, 1.0)
+probe uav 1 (1) rgd(3.5, 1.0)
+probe uav 1 (2) rgd(3.5, 1.0)
+probe uav 1 (3) rgd(3.5, 1.0)
+
+
+[compute shader]
+RWBuffer<float2> u0;
+RWBuffer<double2> u1;
+
+    [numthreads(4, 1, 1)]
+void main(uint id : SV_GroupIndex)
+{
+    double2 d = u0[id];
+    u1[id] = WaveActiveMax(d);
+}
+
+[test]
+dispatch 4 1 1
+probe uav 1 (0) rgd(3.5, 1.5)
+probe uav 1 (1) rgd(3.5, 1.5)
+probe uav 1 (2) rgd(3.5, 1.5)
+probe uav 1 (3) rgd(3.5, 1.5)
+
+
+[compute shader]
+RWBuffer<float2> u0;
+RWBuffer<double2> u1;
+
+    [numthreads(4, 1, 1)]
+void main(uint id : SV_GroupIndex)
+{
+    double2 d = u0[id];
+    u1[id] = WavePrefixSum(d);
+}
+
+[test]
+dispatch 4 1 1
+probe uav 1 (0) rgd(0.0, 0.0)
+probe uav 1 (1) rgd(3.5, 1.0)
+probe uav 1 (2) rgd(7.0, 2.0)
+probe uav 1 (3) rgd(10.5, 3.5)
+
+
+[compute shader]
+RWBuffer<float2> u0;
+RWBuffer<double2> u1;
+
+    [numthreads(4, 1, 1)]
+void main(uint id : SV_GroupIndex)
+{
+    double2 d = u0[id];
+    u1[id] = WavePrefixProduct(d);
+}
+
+[test]
+dispatch 4 1 1
+probe uav 1 (0) rgd(1.0, 1.0)
+probe uav 1 (1) rgd(3.5, 1.0)
+probe uav 1 (2) rgd(12.25, 1.0)
+probe uav 1 (3) rgd(42.875, 1.5)
+
+
+[uav 0]
+format r32g32b32a32-float
+size (buffer, 4)
+
+0.25 0.50 0.75 1.00
+0.50 0.25 1.00 0.75
+0.75 1.00 0.50 0.25
+1.00 0.75 0.25 0.50
+
+[compute shader]
+RWBuffer<float2> u0;
+RWBuffer<double2> u1;
+
+    [numthreads(4, 1, 1)]
+void main(uint id : SV_GroupIndex)
+{
+    uint i = WaveGetLaneIndex() % 4;
+    double2 d = u0[i];
+    u1[i] = QuadReadAcrossX(d);
+}
+
+[test]
+dispatch 4 1 1
+probe uav 1 (0) rgd(0.5, 0.25)
+probe uav 1 (1) rgd(0.25, 0.5)
+probe uav 1 (2) rgd(1.0, 0.75)
+probe uav 1 (3) rgd(0.75, 1.0)
+
+
+[compute shader]
+RWBuffer<float2> u0;
+RWBuffer<double2> u1;
+
+    [numthreads(4, 1, 1)]
+void main(uint id : SV_GroupIndex)
+{
+    uint i = WaveGetLaneIndex() % 4;
+    double2 d = u0[i];
+    u1[i] = QuadReadLaneAt(d, 2);
+}
+
+[test]
+dispatch 4 1 1
+probe uav 1 (0) rgd(0.75, 1.0)
+probe uav 1 (1) rgd(0.75, 1.0)
+probe uav 1 (2) rgd(0.75, 1.0)
+probe uav 1 (3) rgd(0.75, 1.0)

tests/shader_runner.c

@@ -1259,6 +1259,17 @@ static void parse_test_directive(struct shader_runner *runner, const char *line)
             todo_if(runner->is_todo) bug_if(runner->is_bug)
             check_readback_data_uint64(rb, &box, expect, 0);
         }
+        else if (match_string(line, "rgd", &line))
+        {
+            struct dvec2 v;
+
+            ret = sscanf(line, "( %lf , %lf ) %u", &v.x, &v.y, &ulps);
+            if (ret < 2)
+                fatal_error("Malformed probe arguments '%s'.\n", line);
+            if (ret < 3)
+                ulps = 0;
+            todo_if(runner->is_todo) check_readback_data_dvec2(rb, &rect, &v, ulps);
+        }
         else if (match_string(line, "rd", &line))
         {
             double expect;

tests/shader_runner_d3d12.c

@@ -1013,6 +1013,7 @@ static void d3d12_runner_init_caps(struct d3d12_shader_runner *runner,
         DXGI_FORMAT_R16G16B16A16_FLOAT,
         DXGI_FORMAT_R16G16B16A16_UINT,
         DXGI_FORMAT_R16G16B16A16_SINT,
+        DXGI_FORMAT_R32G32_FLOAT,
         DXGI_FORMAT_R8G8B8A8_UNORM,
         DXGI_FORMAT_R8G8B8A8_UINT,
         DXGI_FORMAT_R8G8B8A8_SINT,

tests/utils.h

@@ -222,6 +222,12 @@ static inline bool compare_vec4(const struct vec4 *v1, const struct vec4 *v2, un
     return compare_vec(v1, v2, ulps, 4);
 }
 
+static inline bool compare_dvec2(const struct dvec2 *v1, const struct dvec2 *v2, unsigned int ulps)
+{
+    return compare_double(v1->x, v2->x, ulps)
+            && compare_double(v1->y, v2->y, ulps);
+}
+
 static inline void set_rect(RECT *rect, int left, int top, int right, int bottom)
 {
     rect->left = left;
@@ -262,6 +268,11 @@ static const struct vec4 *get_readback_vec4(const struct resource_readback *rb,
     return get_readback_data(rb, x, y, 0, sizeof(struct vec4));
 }
 
+static const struct dvec2 *get_readback_dvec2(const struct resource_readback *rb, unsigned int x, unsigned int y)
+{
+    return get_readback_data(rb, x, y, 0, sizeof(struct dvec2));
+}
+
 static const struct uvec4 *get_readback_uvec4(const struct resource_readback *rb, unsigned int x, unsigned int y)
 {
     return get_readback_data(rb, x, y, 0, sizeof(struct uvec4));
@@ -419,6 +430,36 @@ static inline void check_readback_data_vec_(const char *file, unsigned int line,
             got.x, got.y, got.z, got.w, expected->x, expected->y, expected->z, expected->w, x, y);
 }
 
+#define check_readback_data_dvec2(a, b, c, d) check_readback_data_dvec2_(__FILE__, __LINE__, a, b, c, d)
+static inline void check_readback_data_dvec2_(const char *file, unsigned int line, const struct resource_readback *rb,
+        const RECT *rect, const struct dvec2 *expected, unsigned int max_diff)
+{
+    RECT r = {0, 0, rb->width, rb->height};
+    unsigned int x = 0, y = 0;
+    struct dvec2 got = {0};
+    bool all_match = true;
+
+    if (rect)
+        r = *rect;
+
+    for (y = r.top; y < r.bottom; ++y)
+    {
+        for (x = r.left; x < r.right; ++x)
+        {
+            got = *get_readback_dvec2(rb, x, y);
+            if (!compare_dvec2(&got, expected, max_diff))
+            {
+                all_match = false;
+                break;
+            }
+        }
+        if (!all_match)
+            break;
+    }
+    ok_(file, line)(all_match, "Got {%.15e, %.15e}, expected {%.15e, %.15e} at (%u, %u).\n",
+            got.x, got.y, expected->x, expected->y, x, y);
+}
+
 #define check_readback_data_ivec4(a, b, c) \
         check_readback_data_uvec4_(__FILE__, __LINE__, a, b, (const struct uvec4 *)(c))
 #define check_readback_data_uvec4(a, b, c) check_readback_data_uvec4_(__FILE__, __LINE__, a, b, c)