fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-29 01:30:08 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

intel: Fix SIMD16 unaligned payload GRF reads on Gen4-5.

Browse source 
When the SIMD16 Gen4-5 fragment shader payload contains source depth
(g2-3), destination stencil (g4), and destination depth (g5-6), the
single register of stencil makes the destination depth unaligned.

We were generating this instruction in the RT write payload setup:

   mov(16)   m14<1>F   g5<8,8,1>F   { align1 compr };

which is illegal, instructions with a source region spanning more than
one register need to be aligned to even registers.  This is because the
hardware implicitly does (nr | 1) instead of (nr + 1) when splitting the
compressed instruction into two mov(8)'s.

I believe this would cause the hardware to load g5 twice, replicating
subspan 0-1's destination depth to subspan 2-3.  This showed up as 2x2
artifact blocks in both TIS-100 and Reicast.

Normally, we rely on the register allocator to even-align our virtual
GRFs.  But we don't control the payload, so we need to lower SIMD widths
to make it work.  To fix this, we teach lower_simd_width about the
restriction, and then call it again after lower_load_payload (which is
what generates the offending MOV).

Fixes: 8aee87fe4c (i965: Use SIMD16 instead of SIMD8 on Gen4 when possible.)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107212
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=13728
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Diego Viola <diego.viola@gmail.com>
This commit is contained in:
Kenneth Graunke 2018-08-02 15:02:18 -07:00
parent 11b9f63a74
commit 08a5c395ab
1 changed files with 20 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
src/intel/compiler/brw_fs.cpp
View file

@ -5115,6 +5115,25 @@ get_fpu_lowered_simd_width(const struct gen_device_info *devinfo,
}
}
if (devinfo->gen < 6) {
/* From the G45 PRM, Volume 4 Page 361:
*
* "Operand Alignment Rule: With the exceptions listed below, a
* source/destination operand in general should be aligned to even
* 256-bit physical register with a region size equal to two 256-bit
* physical registers."
*
* Normally we enforce this by allocating virtual registers to the
* even-aligned class. But we need to handle payload registers.
*/
for (unsigned i = 0; i < inst->sources; i++) {
if (inst->src[i].file == FIXED_GRF && (inst->src[i].nr & 1) &&
inst->size_read(i) > REG_SIZE) {
max_width = MIN2(max_width, 8);
}
}
}
/* From the IVB PRMs:
* "When an instruction is SIMD32, the low 16 bits of the execution mask
* are applied for both halves of the SIMD32 instruction. If different
@ -6321,6 +6340,7 @@ fs_visitor::optimize()
if (OPT(lower_load_payload)) {
split_virtual_grfs();
OPT(register_coalesce);
OPT(lower_simd_width);
OPT(compute_to_mrf);
OPT(dead_code_eliminate);
}