- the whole fragment pipeline is code generated in a single function
+ - input interpolation
+
- depth testing
+ - texture sampling (not all state/formats are supported)
+
- fragment shader TGSI translation
- same level of support as the TGSI SSE2 exec machine, with the exception
we don't fallback to TGSI interpretation when an unsupported opcode is
found, but just ignore it
- - texture sampling via an intrinsic call
- done in SoA layout
- input interpolation also code generated
any width and length
- not all operations are implemented for these types yet though
-Most mesa/progs/demos/* work. Speed is on par with Keith's softpipe-opt branch,
-which includes hand written fast implementations for common cases.
+Most mesa/progs/demos/* work.
To do (probably by this order):
- code generate stipple and stencil testing
- - code generate texture sampling
+ - translate the remaining bits of texture sampling state
- translate TGSI control flow instructions, and all other remaining opcodes
+
+ - integrate with the draw module for VS code generation
- code generate the triangle setup and rasterization
make linux-llvm
-but the rest of these instructions assume scons is used.
+but the rest of these instructions assume that scons is used.
Using
export LD_LIBRARY_PATH=$PWD/build/linux-x86-debug/lib:$LD_LIBRARY_PATH
+For performance evaluation pass debug=no to scons, and use the corresponding
+lib directory without the "-debug" suffix.
+
Unit testing
============
- lp_test_conv: SIMD vector conversion
- lp_test_format: pixel unpacking/packing
-Some of this tests can output results and benchmarks to a tab-seperated-file
+Some of this tests can output results and benchmarks to a tab-separated-file
for posterior analysis, e.g.:
build/linux-x86_64-debug/gallium/drivers/llvmpipe/lp_test_blend -o blend.tsv
at the top of the lp_bld_*.c functions.
- All lp_bld_*.[ch] are isolated from the rest of the driver, and could/may be
- put in a standalone Gallium state -> LLVM IR translation module.
+ put in a stand-alone Gallium state -> LLVM IR translation module.
- We use LLVM-C bindings for now. They are not documented, but follow the C++
interfaces very closely, and appear to be complete enough for code
generation. See
http://npcontemplation.blogspot.com/2008/06/secret-of-llvm-c-bindings.html
- for a standalone example.
+ for a stand-alone example.
int shift = dst_width - n;
res = LLVMBuildShl(builder, res, lp_build_int_const_scalar(src_type, shift), "");
- /* Fill in the empty lower bits for added precision? */
+ /* TODO: Fill in the empty lower bits for additional precision? */
#if 0
{
LLVMValueRef msb;
* Expand the bit width.
*
* This will only change the number of bits the values are represented, not the
- * values themselved.
+ * values themselves.
*/
static void
lp_build_expand(LLVMBuilderRef builder,
LLVMBuildStore(builder, dst, dst_ptr);
}
+ /* FIXME */
assert(!state->occlusion_count);
}
b = LLVMBuildBitCast(bld->builder, b, int_vec_type, "");
}
- /* TODO: On SSE4 we could do this with a single instruction -- PBLENDVB */
-
a = LLVMBuildAnd(bld->builder, a, mask, "");
/* This often gets translated to PANDN, but sometimes the NOT is
return LLVMBuildShuffleVector(bld->builder, a, b, LLVMConstVector(shuffles, n), "");
}
+ else {
#if 0
- else if(0) {
- /* FIXME: Unfortunately select of vectors do not work */
+ /* XXX: Unfortunately select of vectors do not work */
/* Use a select */
LLVMTypeRef elem_type = LLVMInt1Type();
LLVMValueRef cond[LP_MAX_VECTOR_LENGTH];
cond[j + i] = LLVMConstInt(elem_type, cond[i] ? 1 : 0, 0);
return LLVMBuildSelect(bld->builder, LLVMConstVector(cond, n), a, b, "");
- }
-#endif
- else {
+#else
LLVMValueRef mask = lp_build_const_mask_aos(type, cond);
return lp_build_select(bld, mask, a, b);
+#endif
}
}
case PIPE_TEX_WRAP_MIRROR_CLAMP:
case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+ /* FIXME */
default:
assert(0);
}
case PIPE_TEX_FILTER_ANISO:
lp_build_sample_2d_linear_soa(&bld, s, t, width, height, stride, data_ptr, texel);
break;
+ default:
+ assert(0);
}
+ /* FIXME: respect static_state->min_mip_filter */;
+ /* FIXME: respect static_state->mag_img_filter */;
+ /* FIXME: respect static_state->prefilter */;
+
lp_build_sample_compare(&bld, p, texel);
}
break;
case TGSI_UTIL_SIGN_SET:
+ /* TODO: Use bitwese OR for floating point */
res = lp_build_abs( &bld->base, res );
res = LLVMBuildNeg( bld->base.builder, res, "" );
break;
}
-static void
-emit_kilp(
- struct lp_build_tgsi_soa_context *bld )
-{
- /* XXX todo / fix me */
-}
-
-
/**
* Check if inst src/dest regs use indirect addressing into temporary
* register file.
switch (inst->Instruction.Opcode) {
#if 0
case TGSI_OPCODE_ARL:
+ /* FIXME */
FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
tmp0 = emit_fetch( bld, inst, 0, chan_index );
emit_flr(bld, 0, 0);
break;
case TGSI_OPCODE_CND:
+ /* FIXME */
return 0;
break;
return 0;
break;
-#if 0
case TGSI_OPCODE_KILP:
/* predicated kill */
- emit_kilp( bld );
- return 0; /* XXX fix me */
+ /* FIXME */
+ return 0;
break;
-#endif
case TGSI_OPCODE_KIL:
/* conditional kill */
switch( parse.FullToken.Token.Type ) {
case TGSI_TOKEN_TYPE_DECLARATION:
- /* Input already interpolated */
+ /* Inputs already interpolated */
break;
case TGSI_TOKEN_TYPE_INSTRUCTION:
return PIPE_REFERENCED_FOR_WRITE;
}
- /* FIXME: we also need to do the same for the texture cache */
-
return PIPE_UNREFERENCED;
}
struct tgsi_sampler **samplers;
- /* TODO: alpha reference value */
float alpha_ref_value;
- /* TODO: blend constant color */
+ /* FIXME: store (also?) in floats */
uint8_t *blend_color;
struct lp_jit_texture textures[PIPE_MAX_SAMPLERS];
else
depth = NULL;
- /* TODO: blend color */
-
/* XXX: This will most likely fail on 32bit x86 without -mstackrealign */
assert(lp_check_alignment(mask, 16));
projects / mesa.git / commitdiff