VE_COND_MUX_GTE4 is a nice match for the TGSI CMP opcode, however
there is a big limitation due to the general shortcoming of the
vertex shader engine that any instruction can read only two different
temporary registers. So we still have to lower in some cases.
Shader-db RV530:
total instructions in shared programs: 130872 -> 130333 (-0.41%)
instructions in affected programs: 29854 -> 29315 (-1.81%)
helped: 294
HURT: 83
total temps in shared programs: 16747 -> 16775 (0.17%)
temps in affected programs: 407 -> 435 (6.88%)
helped: 10
HURT: 38
Reviewed-by: Filip Gawin <filip.gawin@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23691>
Hidden behind `RUSTICL_ENABLE=fp16` for now as the OpenCL CTS doesn't have
enough fp16 tests at the moment. There has been a lot of work on it though,
so hopefully we can enable and verify it soon.
Additionally libclc also misses a bunch of fp16 functionality, so most of
the tests would also just crash.
However this flag is useful for development as it already wires up most of
the code needed.
Signed-off-by: Karol Herbst <git@karolherbst.de>
Reviewed-by: Nora Allen <blackcatgames@protonmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23788>
so that we don't have to include the structure definitions.
(ac_llvm_compiler includes LLVM, and nir_options includes NIR)
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23517>
This is the compute portion of the work. It uses a separate buffer
for compute SH registers in si_context.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23517>
Instead of writing SH registers into the command buffer, push them into
an array in si_context. Before a draw, take all buffered register writes
and create a single SET_SH_REG_PAIRS_PACKED packet for them.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23517>
so that we don't always disable/enable pipeline stats around blits
when there are no pipeline stat queries
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23517>
The existing code doesn't work with the packed SET packets, so si_pm4_state
needs to find reg_va_low_idx after the whole packet is built.
Remove si_pm4_set_reg_va and do the same thing for SET_SH_REG.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23517>
It can generate all PACKED packets, but only SET_CONTEXT_REG_PAIRS_PACKED
is generated because register shadowing is required by
SET_SH_REG_PAIRS_PACKED*.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23517>
tcs_out_lds_layout is basically renamed to tes_offchip_addr in TCS, using
the same variable as TES and also using the same bit layout. The only
difference in the bit layout was that TCS had to mask out the low bits,
which this also removes.
The enums are renamed to *_SGPR_TCS_OFFCHIP_ADDR so as not to conflict
with *_SGPR_TES_OFFCHIP_ADDR, which are in different user data SGPRs.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23517>
Because a plain mov with the fsat modifier doesn't do a proper 64 bit fsat
we either have to propagate the op as modifier to the instruction that
creates the value, or we add a fake op that applies the fsat op, i.e. we
implement the mov as an add_64 with zero as the second value.
Fixes: 0ff3c4bef2
r600/sfn: drop use of nir source mods
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23754>
