#include "tgsi/tgsi_exec.h"
#include "tgsi/tgsi_dump.h"
-#include "util/u_cpu_detect.h"
+#include "util/u_math.h"
#include "util/u_pointer.h"
#include "util/u_string.h"
elem_types[DRAW_JIT_TEXTURE_DEPTH] = LLVMInt32Type();
elem_types[DRAW_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type();
elem_types[DRAW_JIT_TEXTURE_ROW_STRIDE] =
- LLVMArrayType(LLVMInt32Type(), DRAW_MAX_TEXTURE_LEVELS);
+ LLVMArrayType(LLVMInt32Type(), PIPE_MAX_TEXTURE_LEVELS);
elem_types[DRAW_JIT_TEXTURE_IMG_STRIDE] =
- LLVMArrayType(LLVMInt32Type(), DRAW_MAX_TEXTURE_LEVELS);
+ LLVMArrayType(LLVMInt32Type(), PIPE_MAX_TEXTURE_LEVELS);
elem_types[DRAW_JIT_TEXTURE_DATA] =
LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0),
- DRAW_MAX_TEXTURE_LEVELS);
+ PIPE_MAX_TEXTURE_LEVELS);
elem_types[DRAW_JIT_TEXTURE_MIN_LOD] = LLVMFloatType();
elem_types[DRAW_JIT_TEXTURE_MAX_LOD] = LLVMFloatType();
elem_types[DRAW_JIT_TEXTURE_LOD_BIAS] = LLVMFloatType();
/* struct draw_jit_context */
{
- LLVMTypeRef elem_types[3];
+ LLVMTypeRef elem_types[5];
LLVMTypeRef context_type;
elem_types[0] = LLVMPointerType(LLVMFloatType(), 0); /* vs_constants */
- elem_types[1] = LLVMPointerType(LLVMFloatType(), 0); /* vs_constants */
- elem_types[2] = LLVMArrayType(texture_type,
+ elem_types[1] = LLVMPointerType(LLVMFloatType(), 0); /* gs_constants */
+ elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(LLVMFloatType(), 4), 12), 0); /* planes */
+ elem_types[3] = LLVMPointerType(LLVMFloatType(), 0); /* viewport */
+ elem_types[4] = LLVMArrayType(texture_type,
PIPE_MAX_VERTEX_SAMPLERS); /* textures */
context_type = LLVMStructType(elem_types, Elements(elem_types), 0);
llvm->target, context_type, 0);
LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, gs_constants,
llvm->target, context_type, 1);
+ LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, planes,
+ llvm->target, context_type, 2);
LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures,
llvm->target, context_type,
DRAW_JIT_CTX_TEXTURES);
LLVMAddConstantPropagationPass(llvm->pass);
}
- if(util_cpu_caps.has_sse4_1) {
- /* FIXME: There is a bug in this pass, whereby the combination of fptosi
- * and sitofp (necessary for trunc/floor/ceil/round implementation)
- * somehow becomes invalid code.
- */
- LLVMAddInstructionCombiningPass(llvm->pass);
- }
+ LLVMAddInstructionCombiningPass(llvm->pass);
LLVMAddGVNPass(llvm->pass);
} else {
/* We need at least this pass to prevent the backends to fail in
"instance_divisor");
}
- /* limit index to min(inex, vb_max_index) */
+ /* limit index to min(index, vb_max_index) */
cond = LLVMBuildICmp(builder, LLVMIntULE, index, vb_max_index, "");
index = LLVMBuildSelect(builder, cond, index, vb_max_index, "");
LLVMValueRef clip_ptr0, clip_ptr1, clip_ptr2, clip_ptr3;
LLVMValueRef clip0_ptr, clip1_ptr, clip2_ptr, clip3_ptr;
LLVMValueRef out0elem, out1elem, out2elem, out3elem;
+ int i;
LLVMValueRef ind0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
LLVMValueRef ind1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
clip_ptr2 = draw_jit_header_clip(builder, io2_ptr);
clip_ptr3 = draw_jit_header_clip(builder, io3_ptr);
- for (int i = 0; i<4; i++){
+ for (i = 0; i<4; i++){
clip0_ptr = LLVMBuildGEP(builder, clip_ptr0,
indices, 2, ""); //x0
clip1_ptr = LLVMBuildGEP(builder, clip_ptr1,
}
+/* Equivalent of _mm_set1_ps(a)
+ */
+static LLVMValueRef vec4f_from_scalar(LLVMBuilderRef bld,
+ LLVMValueRef a,
+ const char *name)
+{
+ LLVMValueRef res = LLVMGetUndef(LLVMVectorType(LLVMFloatType(), 4));
+ int i;
+
+ for(i = 0; i < 4; ++i) {
+ LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+ res = LLVMBuildInsertElement(bld, res, a, index, i == 3 ? name : "");
+ }
+
+ return res;
+}
+
/*
* Transforms the outputs for viewport mapping
*/
static void
generate_viewport(struct draw_llvm *llvm,
LLVMBuilderRef builder,
- LLVMValueRef (*outputs)[NUM_CHANNELS])
+ LLVMValueRef (*outputs)[NUM_CHANNELS],
+ LLVMValueRef context_ptr)
{
int i;
- const float *scaleA = llvm->draw->viewport.scale;
- const float *transA = llvm->draw->viewport.translate;
struct lp_type f32_type = lp_type_float_vec(32);
LLVMValueRef out3 = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 w2 w3*/
LLVMValueRef const1 = lp_build_const_vec(f32_type, 1.0); /*1.0 1.0 1.0 1.0*/
-
+ LLVMValueRef vp_ptr = draw_jit_context_viewport(builder, context_ptr);
+
/* for 1/w convention*/
out3 = LLVMBuildFDiv(builder, const1, out3, "");
-
+ LLVMBuildStore(builder, out3, outputs[0][3]);
+
/* Viewport Mapping */
- for (i=0; i<4; i++){
+ for (i=0; i<3; i++){
LLVMValueRef out = LLVMBuildLoad(builder, outputs[0][i], ""); /*x0 x1 x2 x3*/
- LLVMValueRef scale = lp_build_const_vec(f32_type, scaleA[i]); /*sx sx sx sx*/
- LLVMValueRef trans = lp_build_const_vec(f32_type, transA[i]); /*tx tx tx tx*/
+ LLVMValueRef scale;
+ LLVMValueRef trans;
+ LLVMValueRef scale_i;
+ LLVMValueRef trans_i;
+ LLVMValueRef index;
+ index = LLVMConstInt(LLVMInt32Type(), i, 0);
+ scale_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
+
+ index = LLVMConstInt(LLVMInt32Type(), i+4, 0);
+ trans_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
+
+ scale = vec4f_from_scalar(builder, LLVMBuildLoad(builder, scale_i, ""), "scale");
+ trans = vec4f_from_scalar(builder, LLVMBuildLoad(builder, trans_i, ""), "trans");
+
/* divide by w */
- out = LLVMBuildMul(builder, out, out3, "");
+ out = LLVMBuildFMul(builder, out, out3, "");
/* mult by scale */
- out = LLVMBuildMul(builder, out, scale, "");
+ out = LLVMBuildFMul(builder, out, scale, "");
/* add translation */
- out = LLVMBuildAdd(builder, out, trans, "");
+ out = LLVMBuildFAdd(builder, out, trans, "");
/* store transformed outputs */
LLVMBuildStore(builder, out, outputs[0][i]);
}
+
/*
* Returns clipmask as 4xi32 bitmask for the 4 vertices
*/
static LLVMValueRef
generate_clipmask(LLVMBuilderRef builder,
LLVMValueRef (*outputs)[NUM_CHANNELS],
- boolean disable_zclipping,
- boolean enable_d3dclipping)
+ boolean clip_xy,
+ boolean clip_z,
+ boolean clip_user,
+ boolean clip_halfz,
+ unsigned nr,
+ LLVMValueRef context_ptr)
{
LLVMValueRef mask; /* stores the <4xi32> clipmasks */
LLVMValueRef test, temp;
LLVMValueRef zero, shift;
LLVMValueRef pos_x, pos_y, pos_z, pos_w;
+ LLVMValueRef plane1, planes, plane_ptr, sum;
+
+ unsigned i;
struct lp_type f32_type = lp_type_float_vec(32);
+ mask = lp_build_const_int_vec(lp_type_int_vec(32), 0);
+ temp = lp_build_const_int_vec(lp_type_int_vec(32), 0);
zero = lp_build_const_vec(f32_type, 0); /* 0.0f 0.0f 0.0f 0.0f */
shift = lp_build_const_int_vec(lp_type_int_vec(32), 1); /* 1 1 1 1 */
-
+
/* Assuming position stored at output[0] */
pos_x = LLVMBuildLoad(builder, outputs[0][0], ""); /*x0 x1 x2 x3*/
pos_y = LLVMBuildLoad(builder, outputs[0][1], ""); /*y0 y1 y2 y3*/
pos_w = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 w2 w3*/
/* Cliptest, for hardwired planes */
- /* plane 1 */
- test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w);
- temp = shift;
- test = LLVMBuildAnd(builder, test, temp, "");
- mask = test;
+ if (clip_xy){
+ /* plane 1 */
+ test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w);
+ temp = shift;
+ test = LLVMBuildAnd(builder, test, temp, "");
+ mask = test;
- /* plane 2 */
- test = LLVMBuildFAdd(builder, pos_x, pos_w, "");
- test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
- temp = LLVMBuildShl(builder, temp, shift, "");
- test = LLVMBuildAnd(builder, test, temp, "");
- mask = LLVMBuildOr(builder, mask, test, "");
+ /* plane 2 */
+ test = LLVMBuildFAdd(builder, pos_x, pos_w, "");
+ test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+ temp = LLVMBuildShl(builder, temp, shift, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
+ mask = LLVMBuildOr(builder, mask, test, "");
- /* plane 3 */
- test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_y, pos_w);
- temp = LLVMBuildShl(builder, temp, shift, "");
- test = LLVMBuildAnd(builder, test, temp, "");
- mask = LLVMBuildOr(builder, mask, test, "");
-
- /* plane 4 */
- test = LLVMBuildFAdd(builder, pos_y, pos_w, "");
- test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
- temp = LLVMBuildShl(builder, temp, shift, "");
- test = LLVMBuildAnd(builder, test, temp, "");
- mask = LLVMBuildOr(builder, mask, test, "");
-
- if (!disable_zclipping){
- if (enable_d3dclipping){
+ /* plane 3 */
+ test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_y, pos_w);
+ temp = LLVMBuildShl(builder, temp, shift, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
+ mask = LLVMBuildOr(builder, mask, test, "");
+
+ /* plane 4 */
+ test = LLVMBuildFAdd(builder, pos_y, pos_w, "");
+ test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+ temp = LLVMBuildShl(builder, temp, shift, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
+ mask = LLVMBuildOr(builder, mask, test, "");
+ }
+
+ if (clip_z){
+ temp = lp_build_const_int_vec(lp_type_int_vec(32), 16);
+ if (clip_halfz){
/* plane 5 */
test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, pos_z);
- temp = LLVMBuildShl(builder, temp, shift, "");
test = LLVMBuildAnd(builder, test, temp, "");
mask = LLVMBuildOr(builder, mask, test, "");
}
/* plane 5 */
test = LLVMBuildFAdd(builder, pos_z, pos_w, "");
test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
- temp = LLVMBuildShl(builder, temp, shift, "");
test = LLVMBuildAnd(builder, test, temp, "");
mask = LLVMBuildOr(builder, mask, test, "");
}
mask = LLVMBuildOr(builder, mask, test, "");
}
+ if (clip_user){
+ LLVMValueRef planes_ptr = draw_jit_context_planes(builder, context_ptr);
+ LLVMValueRef indices[3];
+ temp = lp_build_const_int_vec(lp_type_int_vec(32), 32);
+
+ /* userclip planes */
+ for (i = 6; i < nr; i++) {
+ indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+ indices[1] = LLVMConstInt(LLVMInt32Type(), i, 0);
+
+ indices[2] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+ plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+ plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_x");
+ planes = vec4f_from_scalar(builder, plane1, "plane4_x");
+ sum = LLVMBuildFMul(builder, planes, pos_x, "");
+
+ indices[2] = LLVMConstInt(LLVMInt32Type(), 1, 0);
+ plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+ plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_y");
+ planes = vec4f_from_scalar(builder, plane1, "plane4_y");
+ test = LLVMBuildFMul(builder, planes, pos_y, "");
+ sum = LLVMBuildFAdd(builder, sum, test, "");
+
+ indices[2] = LLVMConstInt(LLVMInt32Type(), 2, 0);
+ plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+ plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_z");
+ planes = vec4f_from_scalar(builder, plane1, "plane4_z");
+ test = LLVMBuildFMul(builder, planes, pos_z, "");
+ sum = LLVMBuildFAdd(builder, sum, test, "");
+
+ indices[2] = LLVMConstInt(LLVMInt32Type(), 3, 0);
+ plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+ plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_w");
+ planes = vec4f_from_scalar(builder, plane1, "plane4_w");
+ test = LLVMBuildFMul(builder, planes, pos_w, "");
+ sum = LLVMBuildFAdd(builder, sum, test, "");
+
+ test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, sum);
+ temp = LLVMBuildShl(builder, temp, shift, "");
+ test = LLVMBuildAnd(builder, test, temp, "");
+ mask = LLVMBuildOr(builder, mask, test, "");
+ }
+ }
return mask;
}
LLVMValueRef temp;
int i;
- LLVMDumpValue(clipmask);
-
for (i=0; i<4; i++){
temp = LLVMBuildExtractElement(builder, clipmask,
LLVMConstInt(LLVMInt32Type(), i, 0) , "");
ret = LLVMBuildOr(builder, ret, temp, "");
- LLVMDumpValue(ret);
}
LLVMBuildStore(builder, ret, ret_ptr);
- LLVMDumpValue(ret_ptr);
-
}
static void
void *code;
struct lp_build_sampler_soa *sampler = 0;
LLVMValueRef ret, ret_ptr;
- boolean disable_cliptest = variant->key.disable_cliptest;
- boolean disable_viewport = variant->key.disable_viewport;
+ boolean bypass_viewport = variant->key.bypass_viewport;
+ boolean enable_cliptest = variant->key.clip_xy ||
+ variant->key.clip_z ||
+ variant->key.clip_user;
arg_types[0] = llvm->context_ptr_type; /* context */
arg_types[1] = llvm->vertex_header_ptr_type; /* vertex_header */
store_clip(builder, io, outputs);
/* do cliptest */
- if (!disable_cliptest){
+ if (enable_cliptest){
/* allocate clipmask, assign it integer type */
- clipmask = generate_clipmask(builder, outputs,
- variant->key.disable_zclipping, variant->key.enable_d3dclipping);
+ clipmask = generate_clipmask(builder, outputs,
+ variant->key.clip_xy,
+ variant->key.clip_z,
+ variant->key.clip_user,
+ variant->key.clip_halfz,
+ variant->key.nr_planes,
+ context_ptr);
/* return clipping boolean value for function */
clipmask_bool(builder, clipmask, ret_ptr);
}
}
/* do viewport mapping */
- if (!disable_viewport){
- generate_viewport(llvm, builder, outputs);
+ if (!bypass_viewport){
+ generate_viewport(llvm, builder, outputs, context_ptr);
}
/* store clipmask in vertex header and positions in data */
sampler->destroy(sampler);
-#ifdef PIPE_ARCH_X86
- /* Avoid corrupting the FPU stack on 32bit OSes. */
- lp_build_intrinsic(builder, "llvm.x86.mmx.emms", LLVMVoidType(), NULL, 0);
-#endif
-
ret = LLVMBuildLoad(builder, ret_ptr,"");
LLVMBuildRet(builder, ret);
void *code;
struct lp_build_sampler_soa *sampler = 0;
LLVMValueRef ret, ret_ptr;
- boolean disable_cliptest = variant->key.disable_cliptest;
- boolean disable_viewport = variant->key.disable_viewport;
-
+ boolean bypass_viewport = variant->key.bypass_viewport;
+ boolean enable_cliptest = variant->key.clip_xy ||
+ variant->key.clip_z ||
+ variant->key.clip_user;
+
arg_types[0] = llvm->context_ptr_type; /* context */
arg_types[1] = llvm->vertex_header_ptr_type; /* vertex_header */
arg_types[2] = llvm->buffer_ptr_type; /* vbuffers */
store_clip(builder, io, outputs);
/* do cliptest */
- if (!disable_cliptest){
+ if (enable_cliptest){
/* allocate clipmask, assign it integer type */
- clipmask = generate_clipmask(builder, outputs,
- variant->key.disable_zclipping, variant->key.enable_d3dclipping);
+ clipmask = generate_clipmask(builder, outputs,
+ variant->key.clip_xy,
+ variant->key.clip_z,
+ variant->key.clip_user,
+ variant->key.clip_halfz,
+ variant->key.nr_planes,
+ context_ptr);
/* return clipping boolean value for function */
clipmask_bool(builder, clipmask, ret_ptr);
}
}
/* do viewport mapping */
- if (!disable_viewport){
- generate_viewport(llvm, builder, outputs);
+ if (!bypass_viewport){
+ generate_viewport(llvm, builder, outputs, context_ptr);
}
/* store clipmask in vertex header,
sampler->destroy(sampler);
-#ifdef PIPE_ARCH_X86
- /* Avoid corrupting the FPU stack on 32bit OSes. */
- lp_build_intrinsic(builder, "llvm.x86.mmx.emms", LLVMVoidType(), NULL, 0);
-#endif
-
ret = LLVMBuildLoad(builder, ret_ptr,"");
LLVMBuildRet(builder, ret);
key->nr_vertex_elements = llvm->draw->pt.nr_vertex_elements;
/* will have to rig this up properly later */
- key->disable_cliptest = 0;
- key->disable_viewport = 0;
- key->disable_zclipping = 0;
- key->enable_d3dclipping = 0;
+ key->clip_xy = llvm->draw->clip_xy;
+ key->clip_z = llvm->draw->clip_z;
+ key->clip_user = llvm->draw->clip_user;
+ key->bypass_viewport = llvm->draw->identity_viewport;
+ key->clip_halfz = !llvm->draw->rasterizer->gl_rasterization_rules;
+ key->need_edgeflags = (llvm->draw->vs.edgeflag_output ? TRUE : FALSE);
+ key->nr_planes = llvm->draw->nr_planes;
+ key->pad = 0;
/* All variants of this shader will have the same value for
* nr_samplers. Not yet trying to compact away holes in the
unsigned sampler_idx,
uint32_t width, uint32_t height, uint32_t depth,
uint32_t last_level,
- uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS],
- uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS],
- const void *data[DRAW_MAX_TEXTURE_LEVELS])
+ uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
+ uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
+ const void *data[PIPE_MAX_TEXTURE_LEVELS])
{
unsigned j;
struct draw_jit_texture *jit_tex;
}
}
+
+void
+draw_llvm_set_sampler_state(struct draw_context *draw)
+{
+ unsigned i;
+
+ for (i = 0; i < draw->num_samplers; i++) {
+ struct draw_jit_texture *jit_tex = &draw->llvm->jit_context.textures[i];
+
+ if (draw->samplers[i]) {
+ jit_tex->min_lod = draw->samplers[i]->min_lod;
+ jit_tex->max_lod = draw->samplers[i]->max_lod;
+ jit_tex->lod_bias = draw->samplers[i]->lod_bias;
+ COPY_4V(jit_tex->border_color, draw->samplers[i]->border_color);
+ }
+ }
+}
+
+
void
draw_llvm_destroy_variant(struct draw_llvm_variant *variant)
{
projects / mesa.git / blobdiff