summaryrefslogtreecommitdiff
path: root/src/gallium/auxiliary/gallivm
AgeCommit message (Collapse)Author
2010-10-12gallivm: don't branch on KILLs near end of shaderKeith Whitwell
2010-10-11gallivm: More detailed analysis of tgsi shaders.José Fonseca
To allow more optimizations, in particular for direct textures.
2010-10-11gallivm: Eliminate unsigned integer arithmetic from texture coordinates.José Fonseca
SSE support for 32bit and 16bit unsigned arithmetic is not complete, and can easily result in inefficient code. In most cases signed/unsigned doesn't make a difference, such as for integer texture coordinates. So remove uint_coord_type and uint_coord_bld to avoid inefficient operations to sneak in the future.
2010-10-10gallivm: Pass texture coords derivates as scalars.José Fonseca
We end up treating them as scalars in the end, and it saves some instructions.
2010-10-10gallivm: Use variables instead of Phis in loops.José Fonseca
With this commit all explicit Phi emission is now gone.
2010-10-10gallivm: Allow to disable bri-linear filtering with ↵José Fonseca
GALLIVM_DEBUG=no_brilinear runtime option
2010-10-10gallivm: Fix a long standing bug with nested if-then-else emission.José Fonseca
We can't patch true-block at end-if time, as there is no guarantee that the block at the beginning of the true stanza is the same at the end of the true stanza -- other control flow elements may have been emitted half way the true stanza. Although this bug surfaced recently with the commit to skip mip filtering when lod is an integer the bug was always there, although probably it was avoided until now: e.g., cubemap selection nests if-then-else on the else stanza, which does not suffer from the same problem.
2010-10-09gallivm: Cleanup the rest of the flow module.José Fonseca
2010-10-09gallivm: Simplify if/then/else implementation.José Fonseca
No need for for a flow stack anymore.
2010-10-09gallivm: Factor out the SI->FP texture size conversion for SoA path tooJosé Fonseca
2010-10-09gallivm: Remove support for Phi generation.José Fonseca
Simply rely on mem2reg pass. It's easier and more reliable.
2010-10-09gallivm: Use varilables instead of Phis for cubemap selection.José Fonseca
2010-10-09gallivm: Don't generate Phis for execution mask.José Fonseca
2010-10-09gallivm: Special bri-linear computation path for unmodified rho.José Fonseca
2010-10-09gallivm: Less code duplication in log computation.José Fonseca
2010-10-09gallivm: Handle code have ret correctly.José Fonseca
Stop disassembling on unconditional backwards jumps.
2010-10-09llvmpipe: try to be sensible about whether to branch after mask updatesKeith Whitwell
Don't branch more than once in quick succession. Don't branch at the end of the shader.
2010-10-09gallivm: simpler uint8->float conversionsKeith Whitwell
LLVM seems to finds it easier to reason about these than our mantissa-manipulation code.
2010-10-09gallivm: prefer blendvb for integer argumentsKeith Whitwell
2010-10-09llvmpipe: store zero into all alloca'd valuesKeith Whitwell
Fixes slowdown in isosurf with earlier versions of llvm.
2010-10-09gallivm: Do size computations simultanously for all dimensions (AoS).José Fonseca
Operate simultanouesly on <width, height, depth> vector as much as possible, instead of doing the operations on vectors with broadcasted scalars. Also do the 24.8 fixed point scalar with integer shift of the texture size, for unnormalized coordinates. AoS path only for now -- the same thing can be done for SoA.
2010-10-09gallivm: make use of new iround code in lp_bld_conv.Roland Scheidegger
Only requires sse2 now.
2010-10-09gallivm: optimize soa linear clamp to edge wrap mode a bitRoland Scheidegger
Clamp against 0 instead of -0.5, which simplifies things. The former version would have resulted in both int coords being zero (in case of coord being smaller than 0) and some "unused" weight value, whereas now the int coords will be 0 and 1, but weight will be 0, hence the lerp should produce the same value. Still not happy about differences between normalized and non-normalized...
2010-10-09gallivm: avoid unnecessary URem in linear wrap repeat caseRoland Scheidegger
Haven't looked at what code this exactly generates but URem can't be fast. Instead of using two URem only use one and replace the second one with select/add (this is what the corresponding aos code already does).
2010-10-09gallivm: more linear tex wrap mode calculation simplificationRoland Scheidegger
Rearrange order of operations a bit to make some clamps easier. All calculations should be equivalent. Note there seems to be some inconsistency in the clamp to edge case wrt normalized/non-normalized coords, could potentially simplify this too.
2010-10-09gallivm: optimize some tex wrap mode calculations a bitRoland Scheidegger
Sometimes coords are clamped to positive numbers before doing conversion to int, or clamped to 0 afterwards, in this case can use itrunc instead of ifloor which is easier. This is only the case for nearest calculations unfortunately, except linear MIRROR_CLAMP_TO_EDGE which for the same reason can use a unsigned float build context so the ifloor_fract helper can reduce this to itrunc in the ifloor helper itself.
2010-10-09gallivm: replace sub/floor/ifloor combo with ifloor_fractRoland Scheidegger
2010-10-09gallivm: faster iround implementation for sse2Roland Scheidegger
sse2 supports round to nearest directly (or rather, assuming default nearest rounding mode in MXCSR). Use intrinsic to use this rather than round (sse41) or bit manipulation whenever possible.
2010-10-09gallivm: fix trunc/itrunc commentRoland Scheidegger
trunc of -1.5 is -1.0 not 1.0...
2010-10-08gallivm: Remove unnecessary header.Vinson Lee
2010-10-08gallivm: Help for combined extraction and broadcasting.José Fonseca
Doesn't change generated code quality, but saves some typing.
2010-10-08llvmpipe: First minify the texture size, then broadcast.José Fonseca
2010-10-08gallivm: Move into the as much of the second level code as possible.José Fonseca
Also, pass more stuff trhough the sample build context, instead of arguments.
2010-10-08gallivm: Warn when doing inefficient integer comparisons.José Fonseca
2010-10-08gallivm: round rather than truncate in new 4x4f->1x16ub conversion pathKeith Whitwell
2010-10-08gallivm: Use the wrappers for SSE pack intrinsics.José Fonseca
Fixes assertion failures on LLVM 2.6.
2010-10-08gallivm: special case conversion 4x4f to 1x16ubKeith Whitwell
Nice reduction in the number of operations required for final color output in many shaders.
2010-10-08gallivm: Implement brilinear filtering.José Fonseca
2010-10-08gallivm: Fix copy'n'paste typo in previous commit.José Fonseca
2010-10-08gallivm: Clamp mipmap level and zero mip weight simultaneously.José Fonseca
2010-10-08gallivm: Use lp_build_ifloor_fract for lod computation.José Fonseca
Forgot this one before.
2010-10-08gallivm: Don't compute the second mipmap level when frac(lod) == 0José Fonseca
2010-10-08gallivm: Simplify lp_build_mipmap_level_sizes' interface.José Fonseca
2010-10-08gallivm: Do not do mipfiltering when magnifying.José Fonseca
If lod < 0, then invariably follows that ilevel0 == ilevel1 == 0.
2010-10-07gallivm: Vectorize the rho computation.José Fonseca
2010-10-06gallivm: Compute lod as integer whenever possible.José Fonseca
More accurate/faster results for PIPE_TEX_MIPFILTER_NEAREST. Less FP <-> SI conversion overall.
2010-10-06gallivm: Only apply min/max_lod when necessary.José Fonseca
2010-10-06gallivm: don't apply zero lod_biasKeith Whitwell
2010-10-06gallivm: Combined ifloor & fract helper.José Fonseca
The only way to ensure we don't do redundant FP <-> SI conversions.
2010-10-06gallivm: Fast implementation of iround(log2(x))José Fonseca
Not tested yet, but should be correct.