wined3d: texkill ignores the .w only in ps 1.x.

dlls/d3d9/tests/visual.c

@@ -2571,6 +2571,155 @@ static void texdepth_test(IDirect3DDevice9 *device)
     ok(hr == D3D_OK, "IDirect3DDevice9_SetRenderState returned %s\n", DXGetErrorString9(hr));
 }
 
+static void texkill_test(IDirect3DDevice9 *device)
+{
+    IDirect3DPixelShader9 *shader;
+    HRESULT hr;
+    DWORD color;
+
+    const float vertex[] = {
+    /*                          bottom  top    right    left */
+        -1.0,   -1.0,   1.0,   -0.1,    0.9,    0.9,   -0.1,
+         1.0,   -1.0,   0.0,    0.9,   -0.1,    0.9,   -0.1,
+        -1.0,    1.0,   1.0,   -0.1,    0.9,   -0.1,    0.9,
+         1.0,    1.0,   0.0,    0.9,   -0.1,   -0.1,    0.9,
+    };
+
+    DWORD shader_code_11[] = {
+    0xffff0101,                                                             /* ps_1_1                     */
+    0x00000051, 0xa00f0000, 0x3f800000, 0x00000000, 0x00000000, 0x3f800000, /* def c0, 1.0, 0.0, 0.0, 1.0 */
+    0x00000041, 0xb00f0000,                                                 /* texkill t0                 */
+    0x00000001, 0x800f0000, 0xa0e40000,                                     /* mov r0, c0                 */
+    0x0000ffff                                                              /* end                        */
+    };
+    DWORD shader_code_20[] = {
+    0xffff0200,                                                             /* ps_2_0                     */
+    0x0200001f, 0x80000000, 0xb00f0000,                                     /* dcl t0                     */
+    0x05000051, 0xa00f0000, 0x00000000, 0x00000000, 0x3f800000, 0x3f800000, /* def c0, 0.0, 0.0, 1.0, 1.0 */
+    0x01000041, 0xb00f0000,                                                 /* texkill t0                 */
+    0x02000001, 0x800f0800, 0xa0e40000,                                     /* mov oC0, c0                */
+    0x0000ffff                                                              /* end                        */
+    };
+
+    hr = IDirect3DDevice9_Clear(device, 0, NULL, D3DCLEAR_TARGET, 0xff00ff00, 0.0, 0);
+    ok(hr == D3D_OK, "IDirect3DDevice9_CreatePixelShader returned %s\n", DXGetErrorString9(hr));
+    hr = IDirect3DDevice9_CreatePixelShader(device, shader_code_11, &shader);
+    ok(hr == D3D_OK, "IDirect3DDevice9_CreatePixelShader returned %s\n", DXGetErrorString9(hr));
+
+    hr = IDirect3DDevice9_SetPixelShader(device, shader);
+    ok(hr == D3D_OK, "IDirect3DDevice9_SetPixelShader returned %s\n", DXGetErrorString9(hr));
+    hr = IDirect3DDevice9_BeginScene(device);
+    ok(hr == D3D_OK, "IDirect3DDevice9_BeginScene returned %s\n", DXGetErrorString9(hr));
+    if(SUCCEEDED(hr))
+    {
+        hr = IDirect3DDevice9_SetFVF(device, D3DFVF_XYZ | D3DFVF_TEXCOORDSIZE4(0) | D3DFVF_TEX1);
+        ok(hr == D3D_OK, "IDirect3DDevice9_SetFVF returned %s\n", DXGetErrorString9(hr));
+        hr = IDirect3DDevice9_DrawPrimitiveUP(device, D3DPT_TRIANGLESTRIP, 2, vertex, 7 * sizeof(float));
+        ok(hr == D3D_OK, "DrawPrimitiveUP failed (%08x)\n", hr);
+        hr = IDirect3DDevice9_EndScene(device);
+        ok(hr == D3D_OK, "IDirect3DDevice9_EndScene returned %s\n", DXGetErrorString9(hr));
+    }
+    hr = IDirect3DDevice9_Present(device, NULL, NULL, NULL, NULL);
+    ok(hr == D3D_OK, "IDirect3DDevice9_Present failed with %s\n", DXGetErrorString9(hr));
+    color = getPixelColor(device, 63, 46);
+    ok(color == 0x0000ff00, "Pixel 63/46 has color %08x, expected 0x0000ff00\n", color);
+    color = getPixelColor(device, 66, 46);
+    ok(color == 0x0000ff00, "Pixel 66/64 has color %08x, expected 0x0000ff00\n", color);
+    color = getPixelColor(device, 63, 49);
+    ok(color == 0x0000ff00, "Pixel 63/49 has color %08x, expected 0x0000ff00\n", color);
+    color = getPixelColor(device, 66, 49);
+    ok(color == 0x00ff0000, "Pixel 66/49 has color %08x, expected 0x00ff0000\n", color);
+
+    color = getPixelColor(device, 578, 46);
+    ok(color == 0x0000ff00, "Pixel 578/46 has color %08x, expected 0x0000ff00\n", color);
+    color = getPixelColor(device, 575, 46);
+    ok(color == 0x0000ff00, "Pixel 575/64 has color %08x, expected 0x0000ff00\n", color);
+    color = getPixelColor(device, 578, 49);
+    ok(color == 0x0000ff00, "Pixel 578/49 has color %08x, expected 0x0000ff00\n", color);
+    color = getPixelColor(device, 575, 49);
+    ok(color == 0x00ff0000, "Pixel 575/49 has color %08x, expected 0x00ff0000\n", color);
+
+    color = getPixelColor(device, 63, 430);
+    ok(color == 0x0000ff00, "Pixel 578/46 has color %08x, expected 0x0000ff00\n", color);
+    color = getPixelColor(device, 63, 433);
+    ok(color == 0x0000ff00, "Pixel 575/64 has color %08x, expected 0x0000ff00\n", color);
+    color = getPixelColor(device, 66, 433);
+    ok(color == 0x00ff0000, "Pixel 578/49 has color %08x, expected 0x00ff0000\n", color);
+    color = getPixelColor(device, 66, 430);
+    ok(color == 0x00ff0000, "Pixel 575/49 has color %08x, expected 0x00ff0000\n", color);
+
+    color = getPixelColor(device, 578, 430);
+    ok(color == 0x0000ff00, "Pixel 578/46 has color %08x, expected 0x0000ff00\n", color);
+    color = getPixelColor(device, 578, 433);
+    ok(color == 0x0000ff00, "Pixel 575/64 has color %08x, expected 0x0000ff00\n", color);
+    color = getPixelColor(device, 575, 433);
+    ok(color == 0x00ff0000, "Pixel 578/49 has color %08x, expected 0x00ff0000\n", color);
+    color = getPixelColor(device, 575, 430);
+    ok(color == 0x00ff0000, "Pixel 575/49 has color %08x, expected 0x00ff0000\n", color);
+
+    hr = IDirect3DDevice9_SetPixelShader(device, NULL);
+    ok(hr == D3D_OK, "IDirect3DDevice9_SetPixelShader returned %s\n", DXGetErrorString9(hr));
+    IDirect3DPixelShader9_Release(shader);
+
+    hr = IDirect3DDevice9_Clear(device, 0, NULL, D3DCLEAR_TARGET, 0xffffff00, 0.0, 0);
+    ok(hr == D3D_OK, "IDirect3DDevice9_CreatePixelShader returned %s\n", DXGetErrorString9(hr));
+    hr = IDirect3DDevice9_CreatePixelShader(device, shader_code_20, &shader);
+    if(FAILED(hr)) {
+        skip("Failed to create 2.0 test shader, most likely not supported\n");
+        return;
+    }
+
+    hr = IDirect3DDevice9_SetPixelShader(device, shader);
+    ok(hr == D3D_OK, "IDirect3DDevice9_SetPixelShader returned %s\n", DXGetErrorString9(hr));
+    hr = IDirect3DDevice9_BeginScene(device);
+    ok(hr == D3D_OK, "IDirect3DDevice9_BeginScene returned %s\n", DXGetErrorString9(hr));
+    if(SUCCEEDED(hr))
+    {
+        hr = IDirect3DDevice9_DrawPrimitiveUP(device, D3DPT_TRIANGLESTRIP, 2, vertex, 7 * sizeof(float));
+        ok(hr == D3D_OK, "DrawPrimitiveUP failed (%08x)\n", hr);
+        hr = IDirect3DDevice9_EndScene(device);
+        ok(hr == D3D_OK, "IDirect3DDevice9_EndScene returned %s\n", DXGetErrorString9(hr));
+    }
+    hr = IDirect3DDevice9_Present(device, NULL, NULL, NULL, NULL);
+
+    ok(hr == D3D_OK, "IDirect3DDevice9_Present failed with %s\n", DXGetErrorString9(hr));
+    color = getPixelColor(device, 63, 46);
+    ok(color == 0x00ffff00, "Pixel 63/46 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 66, 46);
+    ok(color == 0x00ffff00, "Pixel 66/64 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 63, 49);
+    ok(color == 0x00ffff00, "Pixel 63/49 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 66, 49);
+    ok(color == 0x000000ff, "Pixel 66/49 has color %08x, expected 0x000000ff\n", color);
+
+    color = getPixelColor(device, 578, 46);
+    ok(color == 0x00ffff00, "Pixel 578/46 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 575, 46);
+    ok(color == 0x00ffff00, "Pixel 575/64 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 578, 49);
+    ok(color == 0x00ffff00, "Pixel 578/49 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 575, 49);
+    ok(color == 0x000000ff, "Pixel 575/49 has color %08x, expected 0x000000ff\n", color);
+
+    color = getPixelColor(device, 63, 430);
+    ok(color == 0x00ffff00, "Pixel 578/46 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 63, 433);
+    ok(color == 0x00ffff00, "Pixel 575/64 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 66, 433);
+    ok(color == 0x00ffff00, "Pixel 578/49 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 66, 430);
+    ok(color == 0x000000ff, "Pixel 575/49 has color %08x, expected 0x000000ff\n", color);
+
+    color = getPixelColor(device, 578, 430);
+    ok(color == 0x00ffff00, "Pixel 578/46 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 578, 433);
+    ok(color == 0x00ffff00, "Pixel 575/64 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 575, 433);
+    ok(color == 0x00ffff00, "Pixel 578/49 has color %08x, expected 0x00ffff00\n", color);
+    color = getPixelColor(device, 575, 430);
+    ok(color == 0x000000ff, "Pixel 575/49 has color %08x, expected 0x000000ff\n", color);
+}
+
 START_TEST(visual)
 {
     IDirect3DDevice9 *device_ptr;
@@ -2666,6 +2815,7 @@ START_TEST(visual)
     {
         texbem_test(device_ptr);
         texdepth_test(device_ptr);
+        texkill_test(device_ptr);
     }
     else skip("No ps_1_1 support\n");
 

dlls/wined3d/arb_program_shader.c

@@ -652,8 +652,31 @@ void pshader_hw_map2gl(SHADER_OPCODE_ARG* arg) {
       }
 }
 
-void pshader_hw_tex(SHADER_OPCODE_ARG* arg) {
+void pshader_hw_texkill(SHADER_OPCODE_ARG* arg) {
+    IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
+    DWORD hex_version = This->baseShader.hex_version;
+    SHADER_BUFFER* buffer = arg->buffer;
+    char reg_dest[40];
+
+    /* No swizzles are allowed in d3d's texkill. PS 1.x ignores the 4th component as documented,
+     * but >= 2.0 honors it(undocumented, but tested by the d3d9 testsuit)
+     */
+    pshader_get_register_name(arg->dst, reg_dest);
 
+    if(hex_version >= WINED3DPS_VERSION(2,0)) {
+        /* The arb backend doesn't claim ps 2.0 support, but try to eat what the app feeds to us */
+        shader_addline(buffer, "KIL %s;\n", reg_dest);
+    } else {
+        /* ARB fp doesn't like swizzles on the parameter of the KIL instruction. To mask the 4th component,
+         * copy the register into our general purpose TMP variable, overwrite .w and pass TMP to KIL
+         */
+        shader_addline(buffer, "MOV TMP, %s;\n", reg_dest);
+        shader_addline(buffer, "MOV TMP.w, one.w;\n", reg_dest);
+        shader_addline(buffer, "KIL TMP;\n", reg_dest);
+    }
+}
+
+void pshader_hw_tex(SHADER_OPCODE_ARG* arg) {
     IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
 
     DWORD dst = arg->dst;

dlls/wined3d/glsl_shader.c

@@ -2037,10 +2037,22 @@ void pshader_glsl_texreg2rgb(SHADER_OPCODE_ARG* arg) {
 /** Process the WINED3DSIO_TEXKILL instruction in GLSL.
  * If any of the first 3 components are < 0, discard this pixel */
 void pshader_glsl_texkill(SHADER_OPCODE_ARG* arg) {
+    IWineD3DPixelShaderImpl* This = (IWineD3DPixelShaderImpl*) arg->shader;
+    DWORD hex_version = This->baseShader.hex_version;
     glsl_dst_param_t dst_param;
 
+    /* The argument is a destination parameter, and no writemasks are allowed */
     shader_glsl_add_dst_param(arg, arg->dst, 0, &dst_param);
+    if((hex_version >= WINED3DPS_VERSION(2,0))) {
+        /* 2.0 shaders compare all 4 components in texkill */
+        shader_addline(arg->buffer, "if (any(lessThan(%s.xyzw, vec4(0.0)))) discard;\n", dst_param.reg_name);
+    } else {
+        /* 1.X shaders only compare the first 3 components, propably due to the nature of the texkill
+         * instruction as a tex* instruction, and phase, which kills all a / w components. Even if all
+         * 4 components are defined, only the first 3 are used
+         */
         shader_addline(arg->buffer, "if (any(lessThan(%s.xyz, vec3(0.0)))) discard;\n", dst_param.reg_name);
+    }
 }
 
 /** Process the WINED3DSIO_DP2ADD instruction in GLSL.

dlls/wined3d/pixelshader.c

@@ -218,7 +218,7 @@ CONST SHADER_OPCODE IWineD3DPixelShaderImpl_shader_ins[] = {
     /* Texture */
     {WINED3DSIO_TEXCOORD, "texcoord", "undefined", 1, 1, pshader_hw_texcoord, pshader_glsl_texcoord, 0, WINED3DPS_VERSION(1,3)},
     {WINED3DSIO_TEXCOORD, "texcrd",   "undefined", 1, 2, pshader_hw_texcoord, pshader_glsl_texcoord, WINED3DPS_VERSION(1,4), WINED3DPS_VERSION(1,4)},
-    {WINED3DSIO_TEXKILL,  "texkill",  "KIL",       1, 1, pshader_hw_map2gl, pshader_glsl_texkill, WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(3,0)},
+    {WINED3DSIO_TEXKILL,  "texkill",  "KIL",       1, 1, pshader_hw_texkill, pshader_glsl_texkill, WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(3,0)},
     {WINED3DSIO_TEX,      "tex",      "undefined", 1, 1, pshader_hw_tex, pshader_glsl_tex, 0, WINED3DPS_VERSION(1,3)},
     {WINED3DSIO_TEX,      "texld",    "undefined", 1, 2, pshader_hw_tex, pshader_glsl_tex, WINED3DPS_VERSION(1,4), WINED3DPS_VERSION(1,4)},
     {WINED3DSIO_TEX,      "texld",    "undefined", 1, 3, pshader_hw_tex, pshader_glsl_tex, WINED3DPS_VERSION(2,0), -1},

dlls/wined3d/wined3d_private.h

@@ -1786,6 +1786,7 @@ extern void pshader_hw_texm3x3tex(SHADER_OPCODE_ARG* arg);
 extern void pshader_hw_texm3x3spec(SHADER_OPCODE_ARG* arg);
 extern void pshader_hw_texm3x3vspec(SHADER_OPCODE_ARG* arg);
 extern void pshader_hw_texdepth(SHADER_OPCODE_ARG* arg);
+extern void pshader_hw_texkill(SHADER_OPCODE_ARG* arg);
 
 /* ARB vertex shader prototypes */
 extern void vshader_hw_map2gl(SHADER_OPCODE_ARG* arg);