gallivm: don't use control flow when doing indirect constant buffer lookups

llvm goes crazy when doing that, using way more memory and time, though there's
probably more to it - this points to a very much similar issue as fixed in
8a9f5ecdb116d0449d63f7b94efbfa8b205d826f. In any case I've seen a quite
plain looking vertex shader with just ~50 simple tgsi instructions (but with a
dozen or so such indirect constant buffer lookups) go from a terribly high
~440ms compile time (consuming 25MB of memory in the process) down to a still
awful ~230ms and 13MB with this fix (with llvm 3.3), so there's still obvious
improvements possible (but I have no clue why it's so slow...).
The resulting shader is most likely also faster (certainly seemed so though
I don't have any hard numbers as it may have been influenced by compile times)
since generally fetching constants outside the buffer range is most likely an
app error (that is we expect all indices to be valid).
It is possible this fixes some mysterious vertex shader slowdowns we've seen
ever since we are conforming to newer apis at least partially (the main draw
loop also has similar looking conditionals which we probably could do without -
if not for the fetch at least for the additional elts condition.)

v2: use static vars for the fake bufs, minor code cleanups

Reviewed-by: Jose Fonseca <[email protected]>

1 files changed, 18 insertions, 10 deletions

diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
index 0dfafdca51a..d17d6959b44 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
@@ -271,30 +271,38 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
 static void
 llvm_middle_end_bind_parameters(struct draw_pt_middle_end *middle)
 {
+   static const float fake_const_buf[4];
    struct llvm_middle_end *fpme = llvm_middle_end(middle);
    struct draw_context *draw = fpme->draw;
+   struct draw_llvm *llvm = fpme->llvm;
    unsigned i;
 
-   for (i = 0; i < Elements(fpme->llvm->jit_context.vs_constants); ++i) {
+   for (i = 0; i < Elements(llvm->jit_context.vs_constants); ++i) {
       int num_consts =
          draw->pt.user.vs_constants_size[i] / (sizeof(float) * 4);
-      fpme->llvm->jit_context.vs_constants[i] = draw->pt.user.vs_constants[i];
-      fpme->llvm->jit_context.num_vs_constants[i] = num_consts;
+      llvm->jit_context.vs_constants[i] = draw->pt.user.vs_constants[i];
+      llvm->jit_context.num_vs_constants[i] = num_consts;
+      if (num_consts == 0) {
+         llvm->jit_context.vs_constants[i] = fake_const_buf;
+      }
    }
-   for (i = 0; i < Elements(fpme->llvm->gs_jit_context.constants); ++i) {
+   for (i = 0; i < Elements(llvm->gs_jit_context.constants); ++i) {
       int num_consts =
          draw->pt.user.gs_constants_size[i] / (sizeof(float) * 4);
-      fpme->llvm->gs_jit_context.constants[i] = draw->pt.user.gs_constants[i];
-      fpme->llvm->gs_jit_context.num_constants[i] = num_consts;
+      llvm->gs_jit_context.constants[i] = draw->pt.user.gs_constants[i];
+      llvm->gs_jit_context.num_constants[i] = num_consts;
+      if (num_consts == 0) {
+         llvm->gs_jit_context.constants[i] = fake_const_buf;
+      }
    }
 
-   fpme->llvm->jit_context.planes =
+   llvm->jit_context.planes =
       (float (*)[DRAW_TOTAL_CLIP_PLANES][4]) draw->pt.user.planes[0];
-   fpme->llvm->gs_jit_context.planes =
+   llvm->gs_jit_context.planes =
       (float (*)[DRAW_TOTAL_CLIP_PLANES][4]) draw->pt.user.planes[0];
 
-   fpme->llvm->jit_context.viewports = draw->viewports;
-   fpme->llvm->gs_jit_context.viewports = draw->viewports;
+   llvm->jit_context.viewports = draw->viewports;
+   llvm->gs_jit_context.viewports = draw->viewports;
 }