Module: Mesa Branch: master Commit: d968995c67f5e3a48e49015184f9ea1f143251ec URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=d968995c67f5e3a48e49015184f9ea1f143251ec
Author: Danylo Piliaiev <[email protected]> Date: Mon Dec 21 20:45:42 2020 +0200 turnip: fix SP_HS_WAVE_INPUT_SIZE value It appears that storage for varyings in a wave has an upper limit of wavesize * max_a831 where max_a831 is 64. Exceeding the limit seam to force gpu to reduce primitives processed per wave, at least calculations make sense with such interpretation. With blob SP_HS_WAVE_INPUT_SIZE never exceeds 64 and setting it to 65 in freedreno leads to a hang. Copied from the commit to freedreno e5499ca2 Signed-off-by: Danylo Piliaiev <[email protected]> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8187> --- src/freedreno/vulkan/tu_pipeline.c | 28 +++++++++++++++------------- 1 file changed, 15 insertions(+), 13 deletions(-) diff --git a/src/freedreno/vulkan/tu_pipeline.c b/src/freedreno/vulkan/tu_pipeline.c index c9becdb5dee..3e1b9189df2 100644 --- a/src/freedreno/vulkan/tu_pipeline.c +++ b/src/freedreno/vulkan/tu_pipeline.c @@ -992,7 +992,6 @@ tu6_emit_vpc(struct tu_cs *cs, if (hs) { shader_info *hs_info = &hs->shader->nir->info; - uint32_t unknown_a831 = vs->output_size; tu_cs_emit_pkt4(cs, REG_A6XX_PC_TESS_NUM_VERTEX, 1); tu_cs_emit(cs, hs_info->tess.tcs_vertices_out); @@ -1001,20 +1000,23 @@ tu6_emit_vpc(struct tu_cs *cs, tu_cs_emit_pkt4(cs, REG_A6XX_PC_HS_INPUT_SIZE, 1); tu_cs_emit(cs, patch_control_points * vs->output_size / 4); - /* for A650 this value seems to be local memory size per wave */ - if (vshs_workgroup) { - const uint32_t wavesize = 64; - /* note: if HS is really just the VS extended, then this - * should be by MAX2(patch_control_points, hs_info->tess.tcs_vertices_out) - * however that doesn't match the blob, and fails some dEQP tests. - */ - uint32_t prims_per_wave = wavesize / hs_info->tess.tcs_vertices_out; - uint32_t total_size = vs->output_size * patch_control_points * prims_per_wave; - unknown_a831 = DIV_ROUND_UP(total_size, wavesize); - } + const uint32_t wavesize = 64; + const uint32_t max_wave_input_size = 64; + + /* note: if HS is really just the VS extended, then this + * should be by MAX2(patch_control_points, hs_info->tess.tcs_vertices_out) + * however that doesn't match the blob, and fails some dEQP tests. + */ + uint32_t prims_per_wave = wavesize / hs_info->tess.tcs_vertices_out; + uint32_t max_prims_per_wave = + max_wave_input_size * wavesize / (vs->output_size * patch_control_points); + prims_per_wave = MIN2(prims_per_wave, max_prims_per_wave); + + uint32_t total_size = vs->output_size * patch_control_points * prims_per_wave; + uint32_t wave_input_size = DIV_ROUND_UP(total_size, wavesize); tu_cs_emit_pkt4(cs, REG_A6XX_SP_HS_WAVE_INPUT_SIZE, 1); - tu_cs_emit(cs, unknown_a831); + tu_cs_emit(cs, wave_input_size); /* In SPIR-V generated from GLSL, the tessellation primitive params are * are specified in the tess eval shader, but in SPIR-V generated from _______________________________________________ mesa-commit mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/mesa-commit
