i965: Fix the VS thread limits for GT1, and clarify the WM limits on both.
PATCH 1/1 i965: Fix the VS thread limits for GT1, and clarify the WM limits on both.
---
I don't have GT1 to test with. Does this fix VS regressions for people with that hardware?
src/mesa/drivers/dri/i965/brw_context.c | 13 +++++++++++-- src/mesa/drivers/dri/i965/gen6_vs_state.c | 2 +- src/mesa/drivers/dri/i965/gen6_wm_state.c | 2 +- src/mesa/drivers/dri/intel/intel_chipset.h | 10 ++++++---- 4 files changed, 19 insertions(+), 8 deletions(-)
src/mesa/drivers/dri/i965/brw_context.c +11 -2
@@ -183,8 +183,17 @@ GLboolean brwCreateContext( int api,/* WM maximum threads is number of EUs times number of threads per EU. */ if (intel->gen >=6 ) { brw->urb.size =1024 ; - brw->vs_max_threads =60 ; - brw->wm_max_threads =80 ; +if (IS_GT2 (intel->intelScreen->deviceID)) { +/* This could possibly be 80, but is supposed to require +* disabling of WIZ hashing (bit 6 of GT_MODE, 0x20d0) and a +* GPU reset to change. +*/ + brw->wm_max_threads =40 ; + brw->vs_max_threads =60 ; + }else { + brw->wm_max_threads =40 ; + brw->vs_max_threads =24 ; + } }else if (intel->gen ==5 ) { brw->urb.size =1024 ; brw->vs_max_threads =72 ;
src/mesa/drivers/dri/i965/gen6_vs_state.c +1 -1
@@ -137,7 +137,7 @@ upload_vs_state(struct brw_context *brw) (brw->vs.prog_data->urb_read_length <<GEN6_VS_URB_READ_LENGTH_SHIFT ) | (0 << GEN6_VS_URB_ENTRY_READ_OFFSET_SHIFT)); -OUT_BATCH (((60 -1 ) <<GEN6_VS_MAX_THREADS_SHIFT ) |/* max 60 threads for gen6 */ +OUT_BATCH (((brw->vs_max_threads -1 ) <<GEN6_VS_MAX_THREADS_SHIFT ) |GEN6_VS_STATISTICS_ENABLE |GEN6_VS_ENABLE );ADVANCE_BATCH ();
src/mesa/drivers/dri/i965/gen6_wm_state.c +1 -1
@@ -144,7 +144,7 @@ upload_wm_state(struct brw_context *brw) dw4 |= (brw->wm.prog_data->first_curbe_grf <<GEN6_WM_DISPATCH_START_GRF_SHIFT_0 ); - dw5 |= (40 -1 ) <<GEN6_WM_MAX_THREADS_SHIFT ; + dw5 |= (brw->wm_max_threads -1 ) <<GEN6_WM_MAX_THREADS_SHIFT ;/* CACHE_NEW_WM_PROG */ if (brw->wm.prog_data->dispatch_width ==8 )
src/mesa/drivers/dri/intel/intel_chipset.h +6 -4
@@ -125,12 +125,14 @@/* Compat macro for intel_decode.c */ #define IS_IRONLAKE (devid) IS_GEN5(devid) -#define IS_GEN6 (devid) (devid == PCI_CHIP_SANDYBRIDGE_GT1 || \ - devid == PCI_CHIP_SANDYBRIDGE_GT2 || \ +#define IS_GT2 (devid) (devid == PCI_CHIP_SANDYBRIDGE_GT2 || \ devid == PCI_CHIP_SANDYBRIDGE_GT2_PLUS || \ - devid == PCI_CHIP_SANDYBRIDGE_M_GT1 || \ devid == PCI_CHIP_SANDYBRIDGE_M_GT2 || \ - devid == PCI_CHIP_SANDYBRIDGE_M_GT2_PLUS || \ + devid == PCI_CHIP_SANDYBRIDGE_M_GT2_PLUS) + +#define IS_GEN6 (devid) (IS_GT2(devid) || \ + devid == PCI_CHIP_SANDYBRIDGE_GT1 || \ + devid == PCI_CHIP_SANDYBRIDGE_M_GT1 || \ devid == PCI_CHIP_SANDYBRIDGE_S)#define IS_GT1 (devid) (devid == PCI_CHIP_SANDYBRIDGE_GT1 || \
Signature · 1 line
1.7.4.1
Quoted 4 lines · Eric Anholt
---
I don't have GT1 to test with. Does this fix VS regressions for people with that hardware?
Me neither, but it sure looks correct.
[snip]
Quoted 11 lines
src/mesa/drivers/dri/i965/brw_context.c +1
-2@@ -183,8 +183,17 @@ GLboolean brwCreateContext( int api,/* WM maximum threads is number of EUs times number of threads per EU. */ if (intel->gen>=6 ) { brw->urb.size =1024 ; - brw->vs_max_threads =60 ; - brw->wm_max_threads =80 ; +if (IS_GT2 (intel->intelScreen->deviceID)) {
src/mesa/drivers/dri/i965/brw_context.c GLboolean brwCreateContext( int api, · line 196
Quoted 10 lines
src/mesa/drivers/dri/i965/brw_context.c +10
-0+/* This could possibly be 80, but is supposed to require +* disabling of WIZ hashing (bit 6 of GT_MODE, 0x20d0) and a +* GPU reset to change. +*/ + brw->wm_max_threads =40 ; + brw->vs_max_threads =60 ; + }else { + brw->wm_max_threads =40 ; + brw->vs_max_threads =24 ; + }
Presumably 60 = 12 EUs * 5 threads/EU (for GT2), while 24 = 6 EUs * 4 threads/EU (for GT1). If so, a comment to that effect would be nice...
I still have no idea where 40/80 come from.
[snip]
src/mesa/drivers/dri/intel/intel_chipset.h
Quoted 4 lines
src/mesa/drivers/dri/intel/intel_chipset.h +0
-0
Drop the intel_chipset.h changes, I just committed an equivalent patch.
I had some simple test on that machine.
It fails at VS threads set to 3, but pass at VS thread number 1 and 2. So I believe it's some hidden bug.
One limitation for multi VS thread is
Not to use the scratch space, that will introduce race condition. But I don't think our VS kernel used scratch space.
Thanks
Zou Nanhai
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Eric Anholt
Sent: 2011年3月30日 6:17
To: [email protected]
Subject: [Intel-gfx] [PATCH] i965: Fix the VS thread limits for GT1, and clarify the WM limits on both.
---
I don't have GT1 to test with. Does this fix VS regressions for people with that hardware?
src/mesa/drivers/dri/i965/brw_context.c | 13 +++++++++++-- src/mesa/drivers/dri/i965/gen6_vs_state.c | 2 +- src/mesa/drivers/dri/i965/gen6_wm_state.c | 2 +- src/mesa/drivers/dri/intel/intel_chipset.h | 10 ++++++---- 4 files changed, 19 insertions(+), 8 deletions(-)
src/mesa/drivers/dri/i965/brw_context.c +11 -2
@@ -183,8 +183,17 @@ GLboolean brwCreateContext( int api,/* WM maximum threads is number of EUs times number of threads per EU. */ if (intel->gen >=6 ) { brw->urb.size =1024 ; - brw->vs_max_threads =60 ; - brw->wm_max_threads =80 ; +if (IS_GT2 (intel->intelScreen->deviceID)) { +/* This could possibly be 80, but is supposed to require +* disabling of WIZ hashing (bit 6 of GT_MODE, 0x20d0) and a +* GPU reset to change. +*/ + brw->wm_max_threads =40 ; + brw->vs_max_threads =60 ; + }else { + brw->wm_max_threads =40 ; + brw->vs_max_threads =24 ; + } }else if (intel->gen ==5 ) { brw->urb.size =1024 ; brw->vs_max_threads =72 ;
src/mesa/drivers/dri/i965/gen6_vs_state.c +1 -1
@@ -137,7 +137,7 @@ upload_vs_state(struct brw_context *brw) (brw->vs.prog_data->urb_read_length <<GEN6_VS_URB_READ_LENGTH_SHIFT ) | (0 << GEN6_VS_URB_ENTRY_READ_OFFSET_SHIFT)); -OUT_BATCH (((60 -1 ) <<GEN6_VS_MAX_THREADS_SHIFT ) |/* max 60 threads for gen6 */ +OUT_BATCH (((brw->vs_max_threads -1 ) <<GEN6_VS_MAX_THREADS_SHIFT ) |GEN6_VS_STATISTICS_ENABLE |GEN6_VS_ENABLE );ADVANCE_BATCH ();
src/mesa/drivers/dri/i965/gen6_wm_state.c +1 -1
@@ -144,7 +144,7 @@ upload_wm_state(struct brw_context *brw) dw4 |= (brw->wm.prog_data->first_curbe_grf <<GEN6_WM_DISPATCH_START_GRF_SHIFT_0 ); - dw5 |= (40 -1 ) <<GEN6_WM_MAX_THREADS_SHIFT ; + dw5 |= (brw->wm_max_threads -1 ) <<GEN6_WM_MAX_THREADS_SHIFT ;/* CACHE_NEW_WM_PROG */ if (brw->wm.prog_data->dispatch_width ==8 )
src/mesa/drivers/dri/intel/intel_chipset.h +6 -4
@@ -125,12 +125,14 @@/* Compat macro for intel_decode.c */ #define IS_IRONLAKE (devid) IS_GEN5(devid) -#define IS_GEN6 (devid) (devid == PCI_CHIP_SANDYBRIDGE_GT1 || \ - devid == PCI_CHIP_SANDYBRIDGE_GT2 || \ +#define IS_GT2 (devid) (devid == PCI_CHIP_SANDYBRIDGE_GT2 || \ devid == PCI_CHIP_SANDYBRIDGE_GT2_PLUS || \ - devid == PCI_CHIP_SANDYBRIDGE_M_GT1 || \ devid == PCI_CHIP_SANDYBRIDGE_M_GT2 || \ - devid == PCI_CHIP_SANDYBRIDGE_M_GT2_PLUS || \ + devid == PCI_CHIP_SANDYBRIDGE_M_GT2_PLUS) + +#define IS_GEN6 (devid) (IS_GT2(devid) || \ + devid == PCI_CHIP_SANDYBRIDGE_GT1 || \ + devid == PCI_CHIP_SANDYBRIDGE_M_GT1 || \ devid == PCI_CHIP_SANDYBRIDGE_S)#define IS_GT1 (devid) (devid == PCI_CHIP_SANDYBRIDGE_GT1 || \
Signature · 10 lines
1.7.4.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx