This removes type conversions from 16 bits to 32 bits in the main function
and then back to 16 bits in the epilog.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6622>
Fix defect reported by Coverity.
Sizeof not portable (SIZEOF_MISMATCH)
suspicious_sizeof: Passing argument vec_size * 8UL /* sizeof
(LLVMValueRef *) */ to function __builtin_alloca and then casting
the return value to LLVMValueRef * is suspicious. In this
particular case sizeof (LLVMValueRef *) happens to be equal to
sizeof (LLVMValueRef), but this is not a portable assumption.
Fixes: ca74603b4f ("ac/llvm: add better code for isign")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6682>
These consist of the variations nir_op_{i|u|f}{min|max|med}3 which are either
lowered in the backend (LLVM) anyway or can be recombined by the backend (ACO).
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6421>
Otherwise LLVM does not see the pointers as allowing speculative
loads.
The pipeline-db results are pretty wild, but mostly what is to be
expected from allowing more code movement in LLVM:
Totals from affected shaders:
SGPRS: 157728 -> 168336 (6.73 %)
VGPRS: 158628 -> 158664 (0.02 %)
Spilled SGPRs: 10845 -> 24753 (128.24 %)
Spilled VGPRs: 13 -> 13 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 8 -> 8 (0.00 %) dwords per thread
Code Size: 17189180 -> 17313712 (0.72 %) bytes
LDS: 204 -> 204 (0.00 %) blocks
Max Waves: 5700 -> 5687 (-0.23 %)
Wait states: 0 -> 0 (0.00 %)
This gives some boosts for shaders we can move a descriptor load
outside a loop.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3159>
If buffer or stride is unaligned we use the same trick as on gfx6:
load 1 byte at a time and recompose the output if needed.
This change fixes lots of deqp/glcts tests:
- dEQP-GLES2.functional.draw.random.1, 10, ...
- dEQP-GLES2.functional.vertex_arrays.multiple_attributes.stride.3_float2_0_float2_0_float2_17, ...
- dEQP-GLES2.functional.vertex_arrays.single_attribute.first.byte_first24_offset1_stride2_quads256, ...
- dEQP-GLES2.functional.vertex_arrays.single_attribute.strides.buffer_0_17_byte2_vec4_dynamic_draw_quads_1, ...
- dEQP-GLES31.functional.draw_indirect.random.14, ...
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5502>
16-bit types can be used with MUBUF on GFX6-GFX7.
Fixes: c3e0ba52a0 ("ac/nir: support 16-bit data in buffer_load_format opcodes")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5325>
The compiler should emit s_memtime instead of s_memrealtime for
the subgroup scope. I don't know why this LLVM 9 checks was for
but LLVM 8 also has this amdgcn intrinsic.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5117>
This is a requirement for the shaderResourceMinLod feature which
allows to clamp LOD. This uses all image_sample_*_cl variants.
All dEQP-VK.glsl.texture_functions.texture*clamp.* pass.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4989>
16-bit med3 is only supported on GFX9+.
Fixes dEQP-VK.spirv_assembly.instruction.amd_trinary_minmax.mid3.f16.*.
Fixes: d6a07732c9 ("ac: use llvm.amdgcn.fmed3 intrinsic for nir_op_fmed3")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3962>
Lower 64-bit fmed3 because LLVM doesn't expose an intrinsic.
Fixes dEQP-VK.spirv_assembly.instruction.amd_trinary_minmax.mid3.f64.*.
Fixes: d6a07732c9 ("ac: use llvm.amdgcn.fmed3 intrinsic for nir_op_fmed3")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3962>
Because ac_build_optimization_barrier() overwrites the original
src_type, we have to keep track of it before emitting that barrier.
Otherwise, wrong conversions are expected for pointers or small
bitsizes.
By doing this, we no longer need to do the cast dance in
ac_build_readlane_no_opt_barrier(), it was just necessary for
ac_build_optimization_barrier().
This fixes a bunch of crashes with subgroups related tests when
RADV_DEBUG=checkir is enabled, and it also fixes a compiler crash
with The Surge 2.
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2395
Fixes: 0f45d4dc2b ("ac: add ac_build_readlane without optimization barrier")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3535>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3535>
Previously, when cluster_size was set to 0, it always worked as if
the cluster size was 64. This commit fixes it in wave32 mode by
changing to work as if the cluster size was set to 32.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Split out the logic for exclusive scans into a separate function
that makes clear what it does instead of having this opaque 60
line if.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
This implementation is loosely based on ROCm.
https://github.com/RadeonOpenCompute/ROCm-Device-Libs/blob/master/ockl/src/wfredscan.cl
This fixes dEQP-VK.subgroups.arithmetic.*.subgroupexclusive* on GFX10.
Fixes: 227c29a80d ("amd/common/gfx10: implement scan & reduce operations")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
../src/amd/llvm/ac_llvm_build.c: In function ‘ac_build_canonicalize’:
../src/amd/llvm/ac_llvm_build.c:4567:9: warning: ‘intr’ may be used uninitialized in this function [-Wmaybe-uninitialized]
4567 | return ac_build_intrinsic(ctx, intr, type, params, 1,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
4568 | AC_FUNC_ATTR_READNONE);
| ~~~~~~~~~~~~~~~~~~~~~~
../src/amd/llvm/ac_llvm_build.c:4567:9: warning: ‘type’ may be used uninitialized in this function [-Wmaybe-uninitialized]
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>