fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-22 02:40:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

anv: move L3 config emission to genX_state.c

Browse source 
We're about to reuse this at device initialization.

v2: Handle NULL configs on Gen12+

v3: Handle NULL config in emission helper (Jason)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9534>
This commit is contained in:
Lionel Landwerlin 2021-03-09 16:58:59 +02:00 • committed by Marge Bot
parent 914e7a7f73
commit 581e68bc99
3 changed files with 134 additions and 115 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
src/intel/vulkan/anv_genX.h
View file

@ -71,6 +71,10 @@ void genX(cmd_buffer_emit_hashing_mode)(struct anv_cmd_buffer *cmd_buffer,
void genX(flush_pipeline_select_3d)(struct anv_cmd_buffer *cmd_buffer); void genX(flush_pipeline_select_3d)(struct anv_cmd_buffer *cmd_buffer);
void genX(flush_pipeline_select_gpgpu)(struct anv_cmd_buffer *cmd_buffer); void genX(flush_pipeline_select_gpgpu)(struct anv_cmd_buffer *cmd_buffer);
void genX(emit_l3_config)(struct anv_batch *batch,
const struct anv_device *device,
const struct intel_l3_config *cfg);
void genX(cmd_buffer_config_l3)(struct anv_cmd_buffer *cmd_buffer, void genX(cmd_buffer_config_l3)(struct anv_cmd_buffer *cmd_buffer,
const struct intel_l3_config *cfg); const struct intel_l3_config *cfg);

109
src/intel/vulkan/genX_cmd_buffer.c
View file

@ -1892,8 +1892,6 @@ genX(cmd_buffer_config_l3)(struct anv_cmd_buffer *cmd_buffer,
intel_dump_l3_config(cfg, stderr); intel_dump_l3_config(cfg, stderr);
} }
UNUSED const bool has_slm = cfg->n[INTEL_L3P_SLM];
/* According to the hardware docs, the L3 partitioning can only be changed /* According to the hardware docs, the L3 partitioning can only be changed
* while the pipeline is completely drained and the caches are flushed, * while the pipeline is completely drained and the caches are flushed,
* which involves a first PIPE_CONTROL flush which stalls the pipeline... * which involves a first PIPE_CONTROL flush which stalls the pipeline...
@ -1935,112 +1933,7 @@ genX(cmd_buffer_config_l3)(struct anv_cmd_buffer *cmd_buffer,
pc.CommandStreamerStallEnable = true; pc.CommandStreamerStallEnable = true;
} }
#if GEN_GEN >= 8 genX(emit_l3_config)(&cmd_buffer->batch, cmd_buffer->device, cfg);
assert(!cfg->n[INTEL_L3P_IS] && !cfg->n[INTEL_L3P_C] && !cfg->n[INTEL_L3P_T]);
#if GEN_GEN >= 12
#define L3_ALLOCATION_REG GENX(L3ALLOC)
#define L3_ALLOCATION_REG_num GENX(L3ALLOC_num)
#else
#define L3_ALLOCATION_REG GENX(L3CNTLREG)
#define L3_ALLOCATION_REG_num GENX(L3CNTLREG_num)
#endif
anv_batch_write_reg(&cmd_buffer->batch, L3_ALLOCATION_REG, l3cr) {
#if GEN_GEN < 11
l3cr.SLMEnable = has_slm;
#endif
#if GEN_GEN == 11
/* WA_1406697149: Bit 9 "Error Detection Behavior Control" must be set
* in L3CNTLREG register. The default setting of the bit is not the
* desirable behavior.
*/
l3cr.ErrorDetectionBehaviorControl = true;
l3cr.UseFullWays = true;
#endif
l3cr.URBAllocation = cfg->n[INTEL_L3P_URB];
l3cr.ROAllocation = cfg->n[INTEL_L3P_RO];
l3cr.DCAllocation = cfg->n[INTEL_L3P_DC];
l3cr.AllAllocation = cfg->n[INTEL_L3P_ALL];
}
#else
const bool has_dc = cfg->n[INTEL_L3P_DC] || cfg->n[INTEL_L3P_ALL];
const bool has_is = cfg->n[INTEL_L3P_IS] || cfg->n[INTEL_L3P_RO] ||
cfg->n[INTEL_L3P_ALL];
const bool has_c = cfg->n[INTEL_L3P_C] || cfg->n[INTEL_L3P_RO] ||
cfg->n[INTEL_L3P_ALL];
const bool has_t = cfg->n[INTEL_L3P_T] || cfg->n[INTEL_L3P_RO] ||
cfg->n[INTEL_L3P_ALL];
assert(!cfg->n[INTEL_L3P_ALL]);
/* When enabled SLM only uses a portion of the L3 on half of the banks,
* the matching space on the remaining banks has to be allocated to a
* client (URB for all validated configurations) set to the
* lower-bandwidth 2-bank address hashing mode.
*/
const struct gen_device_info *devinfo = &cmd_buffer->device->info;
const bool urb_low_bw = has_slm && !devinfo->is_baytrail;
assert(!urb_low_bw || cfg->n[INTEL_L3P_URB] == cfg->n[INTEL_L3P_SLM]);
/* Minimum number of ways that can be allocated to the URB. */
const unsigned n0_urb = devinfo->is_baytrail ? 32 : 0;
assert(cfg->n[INTEL_L3P_URB] >= n0_urb);
anv_batch_write_reg(&cmd_buffer->batch, GENX(L3SQCREG1), l3sqc) {
l3sqc.ConvertDC_UC = !has_dc;
l3sqc.ConvertIS_UC = !has_is;
l3sqc.ConvertC_UC = !has_c;
l3sqc.ConvertT_UC = !has_t;
#if GEN_IS_HASWELL
l3sqc.L3SQGeneralPriorityCreditInitialization = SQGPCI_DEFAULT;
#else
l3sqc.L3SQGeneralPriorityCreditInitialization =
devinfo->is_baytrail ? BYT_SQGPCI_DEFAULT : SQGPCI_DEFAULT;
#endif
l3sqc.L3SQHighPriorityCreditInitialization = SQHPCI_DEFAULT;
}
anv_batch_write_reg(&cmd_buffer->batch, GENX(L3CNTLREG2), l3cr2) {
l3cr2.SLMEnable = has_slm;
l3cr2.URBLowBandwidth = urb_low_bw;
l3cr2.URBAllocation = cfg->n[INTEL_L3P_URB] - n0_urb;
#if !GEN_IS_HASWELL
l3cr2.ALLAllocation = cfg->n[INTEL_L3P_ALL];
#endif
l3cr2.ROAllocation = cfg->n[INTEL_L3P_RO];
l3cr2.DCAllocation = cfg->n[INTEL_L3P_DC];
}
anv_batch_write_reg(&cmd_buffer->batch, GENX(L3CNTLREG3), l3cr3) {
l3cr3.ISAllocation = cfg->n[INTEL_L3P_IS];
l3cr3.ISLowBandwidth = 0;
l3cr3.CAllocation = cfg->n[INTEL_L3P_C];
l3cr3.CLowBandwidth = 0;
l3cr3.TAllocation = cfg->n[INTEL_L3P_T];
l3cr3.TLowBandwidth = 0;
}
#if GEN_IS_HASWELL
if (cmd_buffer->device->physical->cmd_parser_version >= 4) {
/* Enable L3 atomics on HSW if we have a DC partition, otherwise keep
* them disabled to avoid crashing the system hard.
*/
anv_batch_write_reg(&cmd_buffer->batch, GENX(SCRATCH1), s1) {
s1.L3AtomicDisable = !has_dc;
}
anv_batch_write_reg(&cmd_buffer->batch, GENX(CHICKEN3), c3) {
c3.L3AtomicDisableMask = true;
c3.L3AtomicDisable = !has_dc;
}
}
#endif
#endif
cmd_buffer->state.current_l3_config = cfg; cmd_buffer->state.current_l3_config = cfg;
} }

136
src/intel/vulkan/genX_state.c
View file

@ -291,14 +291,13 @@ init_render_queue_state(struct anv_queue *queue)
#endif #endif
} }
#if GEN_GEN >= 12 #if GEN_GEN >= 11
/* Starting with GFX version 11, SLM is no longer part of the L3$ config
* so it never changes throughout the lifetime of the VkDevice.
*/
const struct intel_l3_config *cfg = intel_get_default_l3_config(&device->info); const struct intel_l3_config *cfg = intel_get_default_l3_config(&device->info);
if (!cfg) { genX(emit_l3_config)(&batch, device, cfg);
/* Platforms with no configs just setup full-way allocation. */ device->l3_config = cfg;
anv_batch_write_reg(&batch, GENX(L3ALLOC), l3a) {
l3a.L3FullWayAllocationEnable = true;
}
}
#endif #endif
anv_batch_emit(&batch, GENX(MI_BATCH_BUFFER_END), bbe); anv_batch_emit(&batch, GENX(MI_BATCH_BUFFER_END), bbe);
@ -336,6 +335,129 @@ genX(init_device_state)(struct anv_device *device)
return res; return res;
} }
void
genX(emit_l3_config)(struct anv_batch *batch,
const struct anv_device *device,
const struct intel_l3_config *cfg)
{
UNUSED const struct gen_device_info *devinfo = &device->info;
UNUSED const bool has_slm = cfg->n[INTEL_L3P_SLM];
#if GEN_GEN >= 8
#if GEN_GEN >= 12
#define L3_ALLOCATION_REG GENX(L3ALLOC)
#define L3_ALLOCATION_REG_num GENX(L3ALLOC_num)
#else
#define L3_ALLOCATION_REG GENX(L3CNTLREG)
#define L3_ALLOCATION_REG_num GENX(L3CNTLREG_num)
#endif
anv_batch_write_reg(batch, L3_ALLOCATION_REG, l3cr) {
if (cfg == NULL) {
#if GEN_GEN >= 12
l3cr.L3FullWayAllocationEnable = true;
#else
unreachable("Invalid L3$ config");
#endif
} else {
#if GEN_GEN < 11
l3cr.SLMEnable = has_slm;
#endif
#if GEN_GEN == 11
/* WA_1406697149: Bit 9 "Error Detection Behavior Control" must be
* set in L3CNTLREG register. The default setting of the bit is not
* the desirable behavior.
*/
l3cr.ErrorDetectionBehaviorControl = true;
l3cr.UseFullWays = true;
#endif /* GEN_GEN == 11 */
assert(cfg->n[INTEL_L3P_IS] == 0);
assert(cfg->n[INTEL_L3P_C] == 0);
assert(cfg->n[INTEL_L3P_T] == 0);
l3cr.URBAllocation = cfg->n[INTEL_L3P_URB];
l3cr.ROAllocation = cfg->n[INTEL_L3P_RO];
l3cr.DCAllocation = cfg->n[INTEL_L3P_DC];
l3cr.AllAllocation = cfg->n[INTEL_L3P_ALL];
}
}
#else /* GEN_GEN < 8 */
const bool has_dc = cfg->n[INTEL_L3P_DC] || cfg->n[INTEL_L3P_ALL];
const bool has_is = cfg->n[INTEL_L3P_IS] || cfg->n[INTEL_L3P_RO] ||
cfg->n[INTEL_L3P_ALL];
const bool has_c = cfg->n[INTEL_L3P_C] || cfg->n[INTEL_L3P_RO] ||
cfg->n[INTEL_L3P_ALL];
const bool has_t = cfg->n[INTEL_L3P_T] || cfg->n[INTEL_L3P_RO] ||
cfg->n[INTEL_L3P_ALL];
assert(!cfg->n[INTEL_L3P_ALL]);
/* When enabled SLM only uses a portion of the L3 on half of the banks,
* the matching space on the remaining banks has to be allocated to a
* client (URB for all validated configurations) set to the
* lower-bandwidth 2-bank address hashing mode.
*/
const bool urb_low_bw = has_slm && !devinfo->is_baytrail;
assert(!urb_low_bw || cfg->n[INTEL_L3P_URB] == cfg->n[INTEL_L3P_SLM]);
/* Minimum number of ways that can be allocated to the URB. */
const unsigned n0_urb = devinfo->is_baytrail ? 32 : 0;
assert(cfg->n[INTEL_L3P_URB] >= n0_urb);
anv_batch_write_reg(batch, GENX(L3SQCREG1), l3sqc) {
l3sqc.ConvertDC_UC = !has_dc;
l3sqc.ConvertIS_UC = !has_is;
l3sqc.ConvertC_UC = !has_c;
l3sqc.ConvertT_UC = !has_t;
#if GEN_IS_HASWELL
l3sqc.L3SQGeneralPriorityCreditInitialization = SQGPCI_DEFAULT;
#else
l3sqc.L3SQGeneralPriorityCreditInitialization =
devinfo->is_baytrail ? BYT_SQGPCI_DEFAULT : SQGPCI_DEFAULT;
#endif
l3sqc.L3SQHighPriorityCreditInitialization = SQHPCI_DEFAULT;
}
anv_batch_write_reg(batch, GENX(L3CNTLREG2), l3cr2) {
l3cr2.SLMEnable = has_slm;
l3cr2.URBLowBandwidth = urb_low_bw;
l3cr2.URBAllocation = cfg->n[INTEL_L3P_URB] - n0_urb;
#if !GEN_IS_HASWELL
l3cr2.ALLAllocation = cfg->n[INTEL_L3P_ALL];
#endif
l3cr2.ROAllocation = cfg->n[INTEL_L3P_RO];
l3cr2.DCAllocation = cfg->n[INTEL_L3P_DC];
}
anv_batch_write_reg(batch, GENX(L3CNTLREG3), l3cr3) {
l3cr3.ISAllocation = cfg->n[INTEL_L3P_IS];
l3cr3.ISLowBandwidth = 0;
l3cr3.CAllocation = cfg->n[INTEL_L3P_C];
l3cr3.CLowBandwidth = 0;
l3cr3.TAllocation = cfg->n[INTEL_L3P_T];
l3cr3.TLowBandwidth = 0;
}
#if GEN_IS_HASWELL
if (device->physical->cmd_parser_version >= 4) {
/* Enable L3 atomics on HSW if we have a DC partition, otherwise keep
* them disabled to avoid crashing the system hard.
*/
anv_batch_write_reg(batch, GENX(SCRATCH1), s1) {
s1.L3AtomicDisable = !has_dc;
}
anv_batch_write_reg(batch, GENX(CHICKEN3), c3) {
c3.L3AtomicDisableMask = true;
c3.L3AtomicDisable = !has_dc;
}
}
#endif /* GEN_IS_HASWELL */
#endif /* GEN_GEN < 8 */
}
void void
genX(emit_multisample)(struct anv_batch *batch, uint32_t samples, genX(emit_multisample)(struct anv_batch *batch, uint32_t samples,
const VkSampleLocationEXT *locations) const VkSampleLocationEXT *locations)