[PATCH v4 1/3] wined3d: Use 3 component norm for 'nrm' opcode in GLSL backend.
Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=8848 Signed-off-by: Paul Gofman <gofmanp(a)gmail.com> --- v2: no changes v3: - Split off w component handling for zero norm in a separate patch. v4: no changes dlls/wined3d/glsl_shader.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/dlls/wined3d/glsl_shader.c b/dlls/wined3d/glsl_shader.c index c934528981..f2fa18dba8 100644 --- a/dlls/wined3d/glsl_shader.c +++ b/dlls/wined3d/glsl_shader.c @@ -4260,8 +4260,12 @@ static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins) mask_size = shader_glsl_get_write_mask_size(write_mask); shader_glsl_add_src_param(ins, &ins->src[0], write_mask, &src_param); - shader_addline(buffer, "tmp0.x = dot(%s, %s);\n", - src_param.param_str, src_param.param_str); + if (mask_size > 3) + shader_addline(buffer, "tmp0.x = dot(vec3(%s), vec3(%s));\n", + src_param.param_str, src_param.param_str); + else + shader_addline(buffer, "tmp0.x = dot(%s, %s);\n", + src_param.param_str, src_param.param_str); shader_glsl_append_dst(buffer, ins); if (mask_size > 1) -- 2.21.0
Signed-off-by: Paul Gofman <gofmanp(a)gmail.com> --- v2: 1. Encode max float / infinity values in nrm_c0. This should workaround output color clamp problem on testbot's llvmpipe gl renderer, and also reveals the behaviour more reliably for relevant cases. As per my local testing, llvmpipe outputs color 0 for big positive ps color output starting from color value of approximatrely 1e7 (1e6 still outputs 0xff as expected); 2. Refined test revealed broken case on some of the warp drivers, added broken case handling. v3: 1. Encoded Inf value in nrm_c0 so that max_float does not get encoded. 2. Added todo_wine for zero norm and non-zero 'w' cases. v4: no changes dlls/d3d9/tests/visual.c | 255 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 255 insertions(+) diff --git a/dlls/d3d9/tests/visual.c b/dlls/d3d9/tests/visual.c index 1991c0a519..99913e3a73 100644 --- a/dlls/d3d9/tests/visual.c +++ b/dlls/d3d9/tests/visual.c @@ -34,6 +34,7 @@ #define COBJMACROS #include <d3d9.h> #include "wine/test.h" +#include "wine/heap.h" struct vec2 { @@ -25982,6 +25983,259 @@ static void test_sysmem_draw(void) DestroyWindow(window); } +static void test_nrm_instruction(void) +{ + IDirect3DVertexDeclaration9 *vertex_declaration; + IDirect3DVertexShader9 *vertex_shader; + IDirect3DPixelShader9 *pixel_shader; + D3DADAPTER_IDENTIFIER9 identifier; + IDirect3DDevice9 *device; + unsigned int body_size; + IDirect3D9 *d3d; + D3DCOLOR colour; + unsigned int i; + DWORD *ps_code; + ULONG refcount; + D3DCAPS9 caps; + BOOL is_warp; + HWND window; + HRESULT hr; + + static const D3DVERTEXELEMENT9 decl_elements[] = + { + {0, 0, D3DDECLTYPE_FLOAT3, D3DDECLMETHOD_DEFAULT, D3DDECLUSAGE_POSITION, 0}, + {0, 12, D3DDECLTYPE_FLOAT2, D3DDECLMETHOD_DEFAULT, D3DDECLUSAGE_TEXCOORD, 0}, + {0, 20, D3DDECLTYPE_FLOAT4, D3DDECLMETHOD_DEFAULT, D3DDECLUSAGE_TEXCOORD, 1}, + D3DDECL_END() + }; + static const struct + { + struct vec3 position; + struct vec2 t0; + struct vec4 t1; + } + quad[] = + { + {{-1.0f, -1.0f, 0.1f}, {1.0f, 1.0f}, {0.0f, 1.0f, 0.3f, 0.5f}}, + {{ 1.0f, -1.0f, 0.1f}, {1.0f, 1.0f}, {0.0f, 1.0f, 0.3f, 0.5f}}, + {{-1.0f, 1.0f, 0.1f}, {1.0f, 1.0f}, {0.0f, 1.0f, 0.3f, 0.5f}}, + {{ 1.0f, 1.0f, 0.1f}, {1.0f, 1.0f}, {0.0f, 1.0f, 0.3f, 0.5f}}, + }; + + static const DWORD vs_code[] = + { + 0xfffe0300, /* vs_3_0 */ + 0x0200001f, 0x80000000, 0x900f0000, /* dcl_position v0 */ + 0x0200001f, 0x80000005, 0x900f0001, /* dcl_texcoord0 v1 */ + 0x0200001f, 0x80010005, 0x900f0002, /* dcl_texcoord1 v2 */ + 0x0200001f, 0x80000000, 0xe00f0000, /* dcl_position o0 */ + 0x0200001f, 0x80000005, 0xe00f0001, /* dcl_texcoord0 o1 */ + 0x0200001f, 0x80010005, 0xe00f0002, /* dcl_texcoord1 o2 */ + 0x0200001f, 0x80020005, 0xe00f0003, /* dcl_texcoord2 o3 */ + 0x02000001, 0xe00f0000, 0x90e40000, /* mov o0, v0 */ + 0x02000001, 0xe00f0001, 0x90e40001, /* mov o1, v1 */ + 0x02000001, 0xe00f0002, 0x90e40002, /* mov o2, v2 */ + 0x02000024, 0x800f0000, 0x90e40002, /* nrm r0, v2 */ + 0x02000001, 0xe00f0003, 0x80e40000, /* mov o3, r0 */ + 0x0000ffff, + }; + static const DWORD ps_header[] = + { + 0xffff0300, /* ps_3_0 */ + }; + static const DWORD ps_footer[] = + { + 0x02000001, 0x800f0800, 0x80e40000, /* mov oC0, r0 */ + 0x0000ffff, /* end */ + }; + static const DWORD nrm_t0[] = + { + 0x0200001f, 0x80000005, 0x90070001, /* dcl_texcoord0 v0.xyz */ + 0x02000024, 0x800f0000, 0x90e40001, /* nrm r0, v0 */ + }; + static const DWORD nrm_xyz_t0[] = + { + 0x0200001f, 0x80000005, 0x90070000, /* dcl_texcoord0 v0.xyz */ + 0x02000001, 0x800f0000, 0xa0e40000, /* mov r0, c0 */ + 0x02000024, 0x80070000, 0x90e40000, /* nrm r0.xyz, v0 */ + 0x02000001, 0x80040000, 0x80ff0000, /* mov r0.z, r0.w */ + }; + static const DWORD nrm_t1[] = + { + 0x0200001f, 0x80010005, 0x900f0000, /* dcl_texcoord1 v0.xyzw */ + 0x02000024, 0x800f0000, 0x90e40000, /* nrm r0, v0 */ + }; + static const DWORD nrm_vs_t1[] = + { + 0x0200001f, 0x80020005, 0x900f0000, /* dcl_texcoord2 v0.xyzw */ + 0x02000001, 0x800f0000, 0x90e40000, /* mov r0, v0 */ + }; + static const DWORD nrm_t1_xyz[] = + { + 0x0200001f, 0x80010005, 0x900f0000, /* dcl_texcoord1 v0.xyzw */ + 0x02000001, 0x800f0000, 0xa0e40000, /* mov r0, c0 */ + 0x02000024, 0x80070000, 0x90e40000, /* nrm r0.xyz, t1 */ + 0x02000001, 0x80040000, 0x80ff0000, /* mov r0.z, r0.w */ + }; + static const DWORD nrm_t1_xy[] = + { + 0x0200001f, 0x80010005, 0x900f0000, /* dcl_texcoord1 v0.xyzw */ + 0x02000001, 0x800f0000, 0xa0e40000, /* mov r0, c0 */ + 0x02000024, 0x800f0000, 0x90540000, /* nrm r0, v0.xy */ + }; + static const DWORD nrm_xyz_t1_yw[] = + { + 0x0200001f, 0x80010005, 0x900f0000, /* dcl_texcoord1 v0.xyzw */ + 0x02000001, 0x800f0000, 0xa0e40000, /* mov r0, c0 */ + 0x02000024, 0x80070000, 0x90fd0000, /* nrm r0.xyz, v0.yw */ + }; + static const DWORD nrm_t1_dcl_xy[] = + { + 0x0200001f, 0x80010005, 0x90030000, /* dcl_texcoord1 v0.xy */ + 0x02000001, 0x800f0000, 0xa0e40000, /* mov r0, c0 */ + 0x02000024, 0x800f0000, 0x90540000, /* nrm r0, v0 */ + }; + static const DWORD nrm_c0[] = + { + 0x05000051, 0xa00f0001, 0x7f800000, 0xff800000, 0x3e99999a, 0x3ecccccd, + /* def c1, Inf, -Inf, 0.3, 0.4 */ + 0x02000024, 0x800f0000, 0xa0e40000, /* nrm r0, c0 */ + 0x02000001, 0x80040000, 0x80ff0000, /* mov r0.z, r0.w */ + 0x02030029, 0x80aa0000, 0xa0000001, /* if_ge r0.z, c1.x */ + 0x02000001, 0x80040000, 0xa0aa0001, /* mov r0.z, c1.z */ + 0x0000002a, /* else */ + 0x02060029, 0x80aa0000, 0xa0550001, /* if_le r0.z, c1.y */ + 0x02000001, 0x80040000, 0xa0ff0001, /* mov r0.z, c1.w */ + 0x0000002b, /* endif */ + 0x0000002b, /* endif */ + }; + static const DWORD nrm_c0_swizzle[] = + { + 0x02000024, 0x800f0000, 0xa0390000, /* nrm r0, c0.yzwx */ + }; + + static const struct + { + const char *name; + const DWORD *ops; + DWORD body_size; + struct vec4 c0; + D3DCOLOR expected_colour; + BOOL todo; + BOOL warp_broken; + D3DCOLOR expected_warp_broken_colour; + } + tests[] = + { + {"nrm_t0", nrm_t0, ARRAY_SIZE(nrm_t0), {0.0f}, 0x00b4b400}, + {"nrm_xyz_t0", nrm_xyz_t0, ARRAY_SIZE(nrm_xyz_t0), {0.1f, 0.1f, 0.1f, 0.1f}, 0x00b4b419}, + {"nrm_t1", nrm_t1, ARRAY_SIZE(nrm_t1), {0.0f}, 0x0000f449}, + {"nrm_vs_t1", nrm_vs_t1, ARRAY_SIZE(nrm_vs_t1), {0.0f}, 0x0000f449}, + {"nrm_t1_xyz", nrm_t1_xyz, ARRAY_SIZE(nrm_t1_xyz), {0.1f, 0.1f, 0.1f, 0.1f}, 0x0000f419}, + {"nrm_t1_xy", nrm_t1_xy, ARRAY_SIZE(nrm_t1_xy), {0.1f, 0.1f, 0.1f, 0.1f}, 0x0000b4b4}, + {"nrm_xyz_t1_yw", nrm_xyz_t1_yw, ARRAY_SIZE(nrm_xyz_t1_yw), {0.1f, 0.1f, 0.1f, 0.1f}, 0x00d06868}, + {"nrm_t1_dcl_xy", nrm_t1_dcl_xy, ARRAY_SIZE(nrm_t1_dcl_xy), {0.1f, 0.1f, 0.1f, 0.1f}, 0x0000b4b4}, + {"nrm_vec4(0.0, 0.0, 0.0, 0.5)", nrm_c0, ARRAY_SIZE(nrm_c0), {0.0f, 0.0f, 0.0f, 0.5f}, 0x0000004c, + TRUE, TRUE, 0x000000ff}, + {"nrm_vec4(0.0, 0.0, 0.0, -0.5)", nrm_c0, ARRAY_SIZE(nrm_c0), {0.0f, 0.0f, 0.0f, -0.5f}, 0x00000066, + TRUE, TRUE, 0x00000000}, + {"nrm_vec4(0.0, 0.0, 0.0, 0.0)", nrm_c0, ARRAY_SIZE(nrm_c0), {0.0f, 0.0f, 0.0f, 0.0f}, 0x00000000}, + {"nrm_vec4(2.0, 0.0, 0.0, 0.5)", nrm_c0, ARRAY_SIZE(nrm_c0), {2.0f, 0.0f, 0.0f, 0.5f}, 0x00ff0040}, + {"nrm_vec4(1.0)", nrm_c0, ARRAY_SIZE(nrm_c0), {1.0f, 1.0f, 1.0f, 1.0f}, 0x00939393}, + {"nrm_vec4(1e36, 1e36, 1e36, 1e36)", nrm_c0, ARRAY_SIZE(nrm_c0), + {1e36f, 1e36f, 1e36f, 1e36f}, 0x00000000}, /* norm calculation is not overflow safe */ + {"nrm_c0_swizzle", nrm_c0_swizzle, ARRAY_SIZE(nrm_c0_swizzle), {1.0f, 1.0f, 0.0f, 2.0f}, 0x007200e4}, + }; + + window = create_window(); + ok(!!window, "Failed to create a window.\n"); + + d3d = Direct3DCreate9(D3D_SDK_VERSION); + ok(!!d3d, "Failed to create a D3D object.\n"); + if (!(device = create_device(d3d, window, window, TRUE))) + { + skip("Failed to create a D3D device, skipping tests.\n"); + IDirect3D9_Release(d3d); + DestroyWindow(window); + return; + } + + hr = IDirect3DDevice9_GetDeviceCaps(device, &caps); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + if (caps.PixelShaderVersion < D3DPS_VERSION(3, 0) || caps.VertexShaderVersion < D3DVS_VERSION(3, 0)) + { + skip("No shader model 3 support, skipping tests.\n"); + IDirect3DDevice9_Release(device); + IDirect3D9_Release(d3d); + DestroyWindow(window); + return; + } + + hr = IDirect3D9_GetAdapterIdentifier(d3d, D3DADAPTER_DEFAULT, 0, &identifier); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + is_warp = adapter_is_warp(&identifier); + + body_size = 0; + for (i = 0; i < ARRAY_SIZE(tests); ++i) + body_size = max(body_size, tests[i].body_size); + + ps_code = heap_alloc(sizeof(ps_header) + body_size * sizeof(*ps_code) + sizeof(ps_footer)); + memcpy(ps_code, ps_header, sizeof(ps_header)); + + hr = IDirect3DDevice9_SetRenderState(device, D3DRS_ZENABLE, FALSE); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + hr = IDirect3DDevice9_SetRenderState(device, D3DRS_CULLMODE, D3DCULL_NONE); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + hr = IDirect3DDevice9_CreateVertexDeclaration(device, decl_elements, &vertex_declaration); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + hr = IDirect3DDevice9_CreateVertexShader(device, vs_code, &vertex_shader); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + hr = IDirect3DDevice9_SetVertexShader(device, vertex_shader); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + hr = IDirect3DDevice9_SetVertexDeclaration(device, vertex_declaration); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + + for (i = 0; i < ARRAY_SIZE(tests); ++i) + { + memcpy(ps_code + ARRAY_SIZE(ps_header), tests[i].ops, sizeof(*ps_code) * tests[i].body_size); + memcpy(ps_code + ARRAY_SIZE(ps_header) + tests[i].body_size, ps_footer, sizeof(ps_footer)); + + hr = IDirect3DDevice9_CreatePixelShader(device, ps_code, &pixel_shader); + ok(hr == D3D_OK, "Got unexpected hr %#x, test %s.\n", hr, tests[i].name); + hr = IDirect3DDevice9_BeginScene(device); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + hr = IDirect3DDevice9_Clear(device, 0, NULL, D3DCLEAR_TARGET, 0x80808080, 0.0f, 0); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + hr = IDirect3DDevice9_SetPixelShader(device, pixel_shader); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + hr = IDirect3DDevice9_SetPixelShaderConstantF(device, 0, &tests[i].c0.x, 1); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + hr = IDirect3DDevice9_DrawPrimitiveUP(device, D3DPT_TRIANGLESTRIP, 2, quad, sizeof(*quad)); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + hr = IDirect3DDevice9_EndScene(device); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + + colour = getPixelColor(device, 320, 240); + todo_wine_if(tests[i].todo) + ok(color_match(colour, tests[i].expected_colour, 1) || broken(is_warp && tests[i].warp_broken + && color_match(colour, tests[i].expected_warp_broken_colour, 1)), + "test %s, expected 0x%08x, got 0x%08x.\n", + tests[i].name, tests[i].expected_colour, colour); + + IDirect3DPixelShader9_Release(pixel_shader); + } + hr = IDirect3DDevice9_Present(device, NULL, NULL, NULL, NULL); + ok(hr == D3D_OK, "Got unexpected hr %#x.\n", hr); + + heap_free(ps_code); + IDirect3DVertexShader9_Release(vertex_shader); + IDirect3DVertexDeclaration9_Release(vertex_declaration); + refcount = IDirect3DDevice9_Release(device); + ok(!refcount, "Device has %u references left.\n", refcount); + IDirect3D9_Release(d3d); + DestroyWindow(window); +} + START_TEST(visual) { D3DADAPTER_IDENTIFIER9 identifier; @@ -26124,4 +26378,5 @@ START_TEST(visual) test_map_synchronisation(); test_color_vertex(); test_sysmem_draw(); + test_nrm_instruction(); } -- 2.21.0
On Thu, 13 Jun 2019 at 18:27, Paul Gofman <gofmanp(a)gmail.com> wrote:
dlls/d3d9/tests/visual.c | 255 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 255 insertions(+)
Turns out this fails for me on Windows: visual.c:25835: Description string: "AMD Radeon HD 6310 Graphics" ... visual.c:25800: Test failed: test nrm_vec4(0.0, 0.0, 0.0, 0.5), expected 0x0000004c, got 0x000000ff. visual.c:25800: Test failed: test nrm_vec4(0.0, 0.0, 0.0, -0.5), expected 0x00000066, got 0x00000000.
On 6/13/19 17:51, Henri Verbeet wrote:
On Thu, 13 Jun 2019 at 18:27, Paul Gofman <gofmanp(a)gmail.com> wrote:
dlls/d3d9/tests/visual.c | 255 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 255 insertions(+)
Turns out this fails for me on Windows:
visual.c:25835: Description string: "AMD Radeon HD 6310 Graphics" ... visual.c:25800: Test failed: test nrm_vec4(0.0, 0.0, 0.0, 0.5), expected 0x0000004c, got 0x000000ff. visual.c:25800: Test failed: test nrm_vec4(0.0, 0.0, 0.0, -0.5), expected 0x00000066, got 0x00000000.
It worked for me on Nvidia... did you try v2 version of the patch by any chance (https://source.winehq.org/patches/data/166431), is it the same?
Signed-off-by: Paul Gofman <gofmanp(a)gmail.com> --- v3: - output 'w' component unchanged in case of zero norm. v4: - simplify output expression. dlls/wined3d/glsl_shader.c | 12 ++---------- 1 file changed, 2 insertions(+), 10 deletions(-) diff --git a/dlls/wined3d/glsl_shader.c b/dlls/wined3d/glsl_shader.c index f2fa18dba8..f7757bf4a9 100644 --- a/dlls/wined3d/glsl_shader.c +++ b/dlls/wined3d/glsl_shader.c @@ -4268,16 +4268,8 @@ static void shader_glsl_nrm(const struct wined3d_shader_instruction *ins) src_param.param_str, src_param.param_str); shader_glsl_append_dst(buffer, ins); - if (mask_size > 1) - { - shader_addline(buffer, "tmp0.x == 0.0 ? vec%u(0.0) : (%s * inversesqrt(tmp0.x)));\n", - mask_size, src_param.param_str); - } - else - { - shader_addline(buffer, "tmp0.x == 0.0 ? 0.0 : (%s * inversesqrt(tmp0.x)));\n", - src_param.param_str); - } + shader_addline(buffer, "tmp0.x == 0.0 ? %s : (%s * inversesqrt(tmp0.x)));\n", + src_param.param_str, src_param.param_str); } static void shader_glsl_scalar_op(const struct wined3d_shader_instruction *ins) -- 2.21.0
Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com>
participants (3)
-
Henri Verbeet -
Henri Verbeet -
Paul Gofman