[WINED3D 8/9] Improve Sampler support on 2.0 and 3.0 shaders
Ivan Gyurdiev
ivg231 at gmail.com
Mon Jun 12 01:59:16 CDT 2006
- track sampler declarations and store the sampler usage in reg_maps
structure
- store a fake sampler usage for 1.X shaders (defined as 2D sampler)
- re-sync glsl TEX implementation with the ARB one (no idea why they
diverged..)
- use sampler type in new TEX implementation to support 2D, 3D, and Cube
sampling
- change drawprim to bind pixel shader samplers
Additional improvements:
- rename texture limit to texcoord to prevent confusion
- add sampler limit, and use that for samplers - *not* the same as
texcoord above
-------------- next part --------------
---
dlls/wined3d/baseshader.c | 65 ++++++++++++++++++++++++++++++++--------
dlls/wined3d/drawprim.c | 33 ++++++++++++++++++++
dlls/wined3d/glsl_shader.c | 61 +++++++++++++++++++++++++-------------
dlls/wined3d/pixelshader.c | 23 +++++++++-----
dlls/wined3d/vertexshader.c | 10 ++++--
dlls/wined3d/wined3d_private.h | 9 +++++-
6 files changed, 155 insertions(+), 46 deletions(-)
cc3544cee1d98042ae6d6c0b85db85f94a23f7ab
diff --git a/dlls/wined3d/baseshader.c b/dlls/wined3d/baseshader.c
index 7877bcd..7771b79 100644
--- a/dlls/wined3d/baseshader.c
+++ b/dlls/wined3d/baseshader.c
@@ -339,6 +339,8 @@ void shader_get_registers_used(
DWORD regtype = shader_get_regtype(param);
unsigned int regnum = param & D3DSP_REGNUM_MASK;
+ /* Vshader: mark attributes used
+ Pshader: mark 3.0 input registers used, save token */
if (D3DSPR_INPUT == regtype) {
if (!pshader)
@@ -348,21 +350,36 @@ void shader_get_registers_used(
shader_parse_decl_usage(reg_maps->semantics_in, usage, param);
+ /* Vshader: mark 3.0 output registers used, save token */
} else if (D3DSPR_OUTPUT == regtype) {
reg_maps->packed_output[regnum] = 1;
shader_parse_decl_usage(reg_maps->semantics_out, usage, param);
- }
- /* Handle samplers here */
+ /* Save sampler usage token */
+ } else if (D3DSPR_SAMPLER == regtype)
+ reg_maps->samplers[regnum] = usage;
/* Skip definitions (for now) */
} else if (D3DSIO_DEF == curOpcode->opcode) {
pToken += curOpcode->num_params;
- /* Set texture registers, and temporary registers */
+ /* Set texture, address, temporary registers */
} else {
int i, limit;
+ /* Declare 1.X samplers implicitly, based on the destination reg. number */
+ if (D3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) == 1 &&
+ (D3DSIO_TEX == curOpcode->opcode ||
+ D3DSIO_TEXM3x2TEX == curOpcode->opcode ||
+ D3DSIO_TEXM3x3TEX == curOpcode->opcode ||
+ D3DSIO_TEXM3x3SPEC == curOpcode->opcode ||
+ D3DSIO_TEXM3x3VSPEC == curOpcode->opcode)) {
+
+ /* Fake sampler usage, only set reserved bit and ttype */
+ DWORD sampler_code = *pToken & D3DSP_REGNUM_MASK;
+ reg_maps->samplers[sampler_code] = (0x1 << 31) | D3DSTT_2D;
+ }
+
/* This will loop over all the registers and try to
* make a bitmask of the ones we're interested in.
*
@@ -511,7 +528,7 @@ void shader_dump_param(
DWORD regtype = shader_get_regtype(param);
/* There are some minor differences between pixel and vertex shaders */
- BOOL pshader = shader_is_pshader_version(This->baseShader.hex_version);
+ char pshader = shader_is_pshader_version(This->baseShader.hex_version);
/* For one, we'd prefer color components to be shown for pshaders.
* FIXME: use the swizzle function for this */
@@ -678,13 +695,13 @@ void shader_generate_arb_declarations(
shader_addline(buffer, "ADDRESS A%ld;\n", i);
}
- for(i = 0; i < This->baseShader.limits.texture; i++) {
+ for(i = 0; i < This->baseShader.limits.texcoord; i++) {
if (reg_maps->texcoord[i])
shader_addline(buffer,"TEMP T%lu;\n", i);
}
/* Texture coordinate registers must be pre-loaded */
- for (i = 0; i < This->baseShader.limits.texture; i++) {
+ for (i = 0; i < This->baseShader.limits.texcoord; i++) {
if (reg_maps->texcoord[i])
shader_addline(buffer, "MOV T%lu, fragment.texcoord[%lu];\n", i, i);
}
@@ -704,15 +721,36 @@ void shader_generate_glsl_declarations(
IWineD3DBaseShaderImpl* This = (IWineD3DBaseShaderImpl*) iface;
int i;
- FIXME("GLSL not fully implemented yet.\n");
+ /* There are some minor differences between pixel and vertex shaders */
+ char pshader = shader_is_pshader_version(This->baseShader.hex_version);
/* Declare the constants (aka uniforms) */
shader_addline(buffer, "uniform vec4 C[%u];\n", This->baseShader.limits.constant_float);
- /* Declare texture samplers
- * TODO: Make this work for textures other than 2D */
- for (i = 0; i < This->baseShader.limits.texture; i++) {
- shader_addline(buffer, "uniform sampler2D mytex%lu;\n", i);
+ /* Declare texture samplers */
+ for (i = 0; i < This->baseShader.limits.sampler; i++) {
+ if (reg_maps->samplers[i]) {
+
+ const char* prefix = pshader? "psampler": "vsampler";
+
+ DWORD stype = reg_maps->samplers[i] & D3DSP_TEXTURETYPE_MASK;
+ switch (stype) {
+
+ case D3DSTT_2D:
+ shader_addline(buffer, "uniform sampler2D %s%lu;\n", prefix, i);
+ break;
+ case D3DSTT_CUBE:
+ shader_addline(buffer, "uniform samplerCube %s%lu;\n", prefix, i);
+ break;
+ case D3DSTT_VOLUME:
+ shader_addline(buffer, "uniform sampler3D %s%lu;\n", prefix, i);
+ break;
+ default:
+ shader_addline(buffer, "uniform unsupported_sampler %s%lu;\n", prefix, i);
+ FIXME("Unrecognized sampler type: %#lx\n", stype);
+ break;
+ }
+ }
}
/* Declare address variables */
@@ -722,8 +760,9 @@ void shader_generate_glsl_declarations(
}
/* Declare texture coordinate temporaries and initialize them */
- for (i = 0; i < This->baseShader.limits.texture; i++) {
- shader_addline(buffer, "vec4 T%lu = gl_TexCoord[%lu];\n", i, i);
+ for (i = 0; i < This->baseShader.limits.texcoord; i++) {
+ if (reg_maps->texcoord[i])
+ shader_addline(buffer, "vec4 T%lu = gl_TexCoord[%lu];\n", i, i);
}
/* Declare input register temporaries */
diff --git a/dlls/wined3d/drawprim.c b/dlls/wined3d/drawprim.c
index acd708c..a4715e7 100644
--- a/dlls/wined3d/drawprim.c
+++ b/dlls/wined3d/drawprim.c
@@ -1686,6 +1686,36 @@ void drawStridedSoftwareVS(IWineD3DDevic
#endif
+/**
+ * Loads (pixel shader) samplers
+ */
+void drawPrimLoadPSamplersGLSL(
+ IWineD3DDevice* iface) {
+
+ IWineD3DDeviceImpl* This = (IWineD3DDeviceImpl *)iface;
+ GLhandleARB programId = This->stateBlock->shaderPrgId;
+ GLhandleARB name_loc;
+ int i;
+ char sampler_name[20];
+
+ if (programId == 0) {
+ /* No GLSL program set - nothing to do. */
+ return;
+ }
+
+ for (i=0; i< GL_LIMITS(textures); ++i) {
+ if (This->stateBlock->textures[i] != NULL) {
+ snprintf(sampler_name, sizeof(sampler_name), "psampler%d", i);
+ name_loc = GL_EXTCALL(glGetUniformLocationARB(programId, sampler_name));
+ if (name_loc != -1) {
+ TRACE_(d3d_shader)("Loading %s for texture %d\n", sampler_name, i);
+ GL_EXTCALL(glUniform1iARB(name_loc, i));
+ checkGLcall("glUniform1iARB");
+ }
+ }
+ }
+}
+
/**
* Loads floating point constants (aka uniforms) into the currently set GLSL program.
* When @constants_set == NULL, it will load all the constants.
@@ -1780,6 +1810,9 @@ void drawPrimLoadConstants(IWineD3DDevic
}
if (usePixelShader) {
+ /* Load pixel shader samplers */
+ drawPrimLoadPSamplersGLSL(iface);
+
/* Load DirectX 9 float constants/uniforms for pixel shader */
drawPrimLoadConstantsGLSL_F(iface, WINED3D_PSHADER_MAX_CONSTANTS,
This->stateBlock->pixelShaderConstantF,
diff --git a/dlls/wined3d/glsl_shader.c b/dlls/wined3d/glsl_shader.c
index bed5b98..32b7f9c 100644
--- a/dlls/wined3d/glsl_shader.c
+++ b/dlls/wined3d/glsl_shader.c
@@ -242,7 +242,10 @@ static void shader_glsl_get_register_nam
}
break;
case D3DSPR_SAMPLER:
- sprintf(tmpStr, "mytex%lu", reg);
+ if (pshader)
+ sprintf(tmpStr, "psampler%lu", reg);
+ else
+ sprintf(tmpStr, "vsampler%lu", reg);
break;
case D3DSPR_COLOROUT:
if (reg == 0)
@@ -729,30 +732,48 @@ void pshader_glsl_tex(SHADER_OPCODE_ARG*
DWORD hex_version = This->baseShader.hex_version;
char dst_str[100], dst_reg[50], dst_mask[6];
- char src0_str[100], src0_reg[50], src0_mask[6];
- char src1_str[100], src1_reg[50], src1_mask[6];
+ char coord_str[100], coord_reg[50], coord_mask[6];
+ char sampler_str[100], sampler_reg[50], sampler_mask[6];
DWORD reg_dest_code = arg->dst & D3DSP_REGNUM_MASK;
+ DWORD sampler_code, sampler_type;
/* All versions have a destination register */
shader_glsl_add_param(arg, arg->dst, 0, FALSE, dst_reg, dst_mask, dst_str);
-
+
/* 1.0-1.3: Use destination register as coordinate source.
+ 1.4+: Use provided coordinate source register. */
+ if (hex_version < D3DPS_VERSION(1,4))
+ strcpy(coord_reg, dst_reg);
+ else
+ shader_glsl_add_param(arg, arg->src[0], arg->src_addr[0], TRUE, coord_reg, coord_mask, coord_str);
+
+ /* 1.0-1.4: Use destination register as coordinate source.
* 2.0+: Use provided coordinate source register. */
- if (hex_version == D3DPS_VERSION(1,4)) {
- shader_glsl_add_param(arg, arg->src[0], arg->src_addr[0], TRUE, src0_reg, src0_mask, src0_str);
- sprintf(src1_str, "mytex%lu", reg_dest_code);
- } else if (hex_version > D3DPS_VERSION(1,4)) {
- shader_glsl_add_param(arg, arg->src[0], arg->src_addr[0], TRUE, src0_reg, src0_mask, src0_str);
- shader_glsl_add_param(arg, arg->src[1], arg->src_addr[1], TRUE, src1_reg, src1_mask, src1_str);
- }
-
- /* 1.0-1.4: Use destination register number as texture code.
- * 2.0+: Use provided sampler number as texure code. */
- if (hex_version < D3DPS_VERSION(1,4)) {
- shader_addline(buffer, "%s = texture2D(mytex%lu, gl_TexCoord[%lu].st);\n",
- dst_str, reg_dest_code, reg_dest_code);
- } else {
- shader_addline(buffer, "%s = texture2D(%s, %s.st);\n", dst_str, src1_str, src0_reg);
+ if (hex_version < D3DPS_VERSION(2,0)) {
+ sprintf(sampler_str, "psampler%lu", reg_dest_code);
+ sampler_code = reg_dest_code;
+ }
+ else {
+ shader_glsl_add_param(arg, arg->src[1], arg->src_addr[1], TRUE, sampler_reg, sampler_mask, sampler_str);
+ sampler_code = arg->src[1] & D3DSP_REGNUM_MASK;
+ }
+
+ sampler_type = arg->reg_maps->samplers[sampler_code] & D3DSP_TEXTURETYPE_MASK;
+ switch(sampler_type) {
+
+ case D3DSTT_2D:
+ shader_addline(buffer, "%s = texture2D(%s, %s.st);\n", dst_str, sampler_str, coord_reg);
+ break;
+ case D3DSTT_CUBE:
+ shader_addline(buffer, "%s = textureCube(%s, %s.stp);\n", dst_str, sampler_str, coord_reg);
+ break;
+ case D3DSTT_VOLUME:
+ shader_addline(buffer, "%s = texture3D(%s, %s.stp);\n", dst_str, sampler_str, coord_reg);
+ break;
+ default:
+ shader_addline(buffer, "%s = unrecognized_stype(%s, %s.stp);\n", dst_str, sampler_str, coord_reg);
+ FIXME("Unrecognized sampler type: %#lx;\n", sampler_type);
+ break;
}
}
@@ -806,7 +827,7 @@ void pshader_glsl_texm3x2tex(SHADER_OPCO
shader_glsl_add_param(arg, arg->src[0], arg->src_addr[0], TRUE, src0_name, src0_mask, src0_str);
shader_addline(buffer, "tmp0.y = dot(vec3(T%lu), vec3(%s));\n", reg, src0_str);
- shader_addline(buffer, "T%lu = texture2D(mytex%lu, tmp0.st);\n", reg, reg);
+ shader_addline(buffer, "T%lu = texture2D(psampler%lu, tmp0.st);\n", reg, reg);
}
void pshader_glsl_input_pack(
diff --git a/dlls/wined3d/pixelshader.c b/dlls/wined3d/pixelshader.c
index 9b83e62..f4fe007 100644
--- a/dlls/wined3d/pixelshader.c
+++ b/dlls/wined3d/pixelshader.c
@@ -947,7 +947,8 @@ static void pshader_set_limits(
This->baseShader.limits.constant_float = 8;
This->baseShader.limits.constant_int = 0;
This->baseShader.limits.constant_bool = 0;
- This->baseShader.limits.texture = 4;
+ This->baseShader.limits.texcoord = 4;
+ This->baseShader.limits.sampler = 4;
This->baseShader.limits.packed_input = 0;
break;
@@ -956,7 +957,8 @@ static void pshader_set_limits(
This->baseShader.limits.constant_float = 8;
This->baseShader.limits.constant_int = 0;
This->baseShader.limits.constant_bool = 0;
- This->baseShader.limits.texture = 6;
+ This->baseShader.limits.texcoord = 6;
+ This->baseShader.limits.sampler = 6;
This->baseShader.limits.packed_input = 0;
break;
@@ -967,7 +969,8 @@ static void pshader_set_limits(
This->baseShader.limits.constant_float = 32;
This->baseShader.limits.constant_int = 16;
This->baseShader.limits.constant_bool = 16;
- This->baseShader.limits.texture = 8;
+ This->baseShader.limits.texcoord = 8;
+ This->baseShader.limits.sampler = 16;
This->baseShader.limits.packed_input = 0;
break;
@@ -976,15 +979,17 @@ static void pshader_set_limits(
This->baseShader.limits.constant_float = 224;
This->baseShader.limits.constant_int = 16;
This->baseShader.limits.constant_bool = 16;
- This->baseShader.limits.texture = 0;
+ This->baseShader.limits.texcoord = 0;
+ This->baseShader.limits.sampler = 16;
This->baseShader.limits.packed_input = 12;
break;
default: This->baseShader.limits.temporary = 32;
- This->baseShader.limits.constant_float = 8;
- This->baseShader.limits.constant_int = 0;
- This->baseShader.limits.constant_bool = 0;
- This->baseShader.limits.texture = 8;
+ This->baseShader.limits.constant_float = 32;
+ This->baseShader.limits.constant_int = 16;
+ This->baseShader.limits.constant_bool = 16;
+ This->baseShader.limits.texcoord = 8;
+ This->baseShader.limits.sampler = 16;
This->baseShader.limits.packed_input = 0;
FIXME("Unrecognized pixel shader version %#lx\n",
This->baseShader.hex_version);
@@ -1124,7 +1129,7 @@ static void pshader_hw_tex(SHADER_OPCODE
get_register_name(dst, reg_dest, arg->reg_maps->constantsF);
/* 1.0-1.3: Use destination register as coordinate source.
- 2.0+: Use provided coordinate source register. */
+ 1.4+: Use provided coordinate source register. */
if (hex_version < D3DPS_VERSION(1,4))
strcpy(reg_coord, reg_dest);
else
diff --git a/dlls/wined3d/vertexshader.c b/dlls/wined3d/vertexshader.c
index 04a8337..ec04d54 100644
--- a/dlls/wined3d/vertexshader.c
+++ b/dlls/wined3d/vertexshader.c
@@ -712,7 +712,7 @@ inline static void vshader_program_add_p
static void vshader_set_limits(
IWineD3DVertexShaderImpl *This) {
- This->baseShader.limits.texture = 0;
+ This->baseShader.limits.texcoord = 0;
This->baseShader.limits.attributes = 16;
This->baseShader.limits.packed_input = 0;
@@ -727,6 +727,7 @@ static void vshader_set_limits(
This->baseShader.limits.constant_int = 0;
This->baseShader.limits.address = 1;
This->baseShader.limits.packed_output = 0;
+ This->baseShader.limits.sampler = 0;
break;
case D3DVS_VERSION(2,0):
@@ -736,6 +737,7 @@ static void vshader_set_limits(
This->baseShader.limits.constant_int = 16;
This->baseShader.limits.address = 1;
This->baseShader.limits.packed_output = 0;
+ This->baseShader.limits.sampler = 0;
break;
case D3DVS_VERSION(3,0):
@@ -744,13 +746,15 @@ static void vshader_set_limits(
This->baseShader.limits.constant_int = 32;
This->baseShader.limits.address = 1;
This->baseShader.limits.packed_output = 12;
+ This->baseShader.limits.sampler = 4;
break;
default: This->baseShader.limits.temporary = 12;
- This->baseShader.limits.constant_bool = 0;
- This->baseShader.limits.constant_int = 0;
+ This->baseShader.limits.constant_bool = 16;
+ This->baseShader.limits.constant_int = 16;
This->baseShader.limits.address = 1;
This->baseShader.limits.packed_output = 0;
+ This->baseShader.limits.sampler = 0;
FIXME("Unrecognized vertex shader version %#lx\n",
This->baseShader.hex_version);
}
diff --git a/dlls/wined3d/wined3d_private.h b/dlls/wined3d/wined3d_private.h
index a53bc1c..5f97fb3 100644
--- a/dlls/wined3d/wined3d_private.h
+++ b/dlls/wined3d/wined3d_private.h
@@ -1269,9 +1269,15 @@ typedef struct shader_reg_maps {
char constantsF[MAX_CONST_F]; /* pixel, vertex */
/* TODO: Integer and bool constants */
+ /* Semantics maps (semantic -> reg_token)
+ * Use 0 as default (bit 31 is always 1 on a valid token) */
DWORD* semantics_in; /* vertex, pixel */
DWORD* semantics_out; /* vertex */
+ /* Sampler usage tokens
+ * Use 0 as default (bit 31 is always 1 on a valid token) */
+ DWORD samplers[MAX_SAMPLERS];
+
} shader_reg_maps;
#define SHADER_PGMSIZE 65535
@@ -1307,7 +1313,8 @@ typedef struct SHADER_OPCODE_ARG {
typedef struct SHADER_LIMITS {
unsigned int temporary;
- unsigned int texture;
+ unsigned int texcoord;
+ unsigned int sampler;
unsigned int constant_int;
unsigned int constant_float;
unsigned int constant_bool;
--
1.3.3
More information about the wine-patches
mailing list