Igt-dev Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: "Grzegorzek, Dominik" <dominik.grzegorzek@intel.com>
To: "igt-dev@lists.freedesktop.org" <igt-dev@lists.freedesktop.org>,
	"Hajda, Andrzej" <andrzej.hajda@intel.com>
Cc: "Kempczynski, Zbigniew" <zbigniew.kempczynski@intel.com>,
	"Mun, Gwan-gyeong" <gwan-gyeong.mun@intel.com>,
	"Manszewski, Christoph" <christoph.manszewski@intel.com>,
	"kamil.konieczny@linux.intel.com"
	<kamil.konieczny@linux.intel.com>
Subject: Re: [PATCH v4 4/4] lib/gpgpu_shader: add support for Xe3 platforms
Date: Mon, 25 Nov 2024 14:08:09 +0000	[thread overview]
Message-ID: <7f1698d519d08d158043339681300cee95da2b0e.camel@intel.com> (raw)
In-Reply-To: <20241125-gpgpu_send_rework-v4-4-c16b568a1f3d@intel.com>

On Mon, 2024-11-25 at 08:31 +0100, Andrzej Hajda wrote:
> > Xe3 platforms disallow indirect load/store addressing. Surface descriptor
> > must be passed in 2DBlock payload. Use for it inline data passed from
> > thread dispatcher.

To be clear, mention that system routine is the entity which adds
this constrain.
> > 
> > v4:
> >   - moved gpgpu_alloc_gpu_addr changes to previous patch (Dominik)
> > 
> > Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>
> > ---
> >  lib/iga64_generated_codes.c | 13 ++++++++-----
> >  lib/iga64_macros.h          | 37 +++++++++++++++++++++++++++++++++----
> >  2 files changed, 41 insertions(+), 9 deletions(-)
> > 
> > diff --git a/lib/iga64_generated_codes.c b/lib/iga64_generated_codes.c
> > index 04015b0b6d29..a12135e7dbfd 100644
> > --- a/lib/iga64_generated_codes.c
> > +++ b/lib/iga64_generated_codes.c
> > @@ -3,7 +3,7 @@
> >  
> >  #include "gpgpu_shader.h"
> >  
> > -#define MD5_SUM_IGA64_ASMS da66be3cf9bec819a61429de75943011
> > +#define MD5_SUM_IGA64_ASMS 7b1db60d1de46cf35666f2a7f51e8fc2
> >  
> >  struct iga64_template const iga64_code_gpgpu_fill[] = {
> >  	{ .gen_ver = 2000, .size = 44, .code = (const uint32_t []) {
> > @@ -80,10 +80,11 @@ struct iga64_template const iga64_code_gpgpu_fill[] = {
> >  };
> >  
> >  struct iga64_template const iga64_code_end_system_routine_step_if_eq[] = {
> > -	{ .gen_ver = 2000, .size = 44, .code = (const uint32_t []) {
> > +	{ .gen_ver = 2000, .size = 48, .code = (const uint32_t []) {
> >  		0x80000966, 0x80018220, 0x02008000, 0x00008000,
> >  		0x80000965, 0x80118220, 0x02008010, 0xc0ded000,
> >  		0x800c0961, 0x1e054220, 0x00000000, 0x00000000,
> > +		0x80000061, 0x1e554220, 0x00000000, 0x00000000,
> >  		0x80000061, 0x1e654220, 0x00000000, 0xc0ded000,
> >  		0x80000061, 0x1e754220, 0x00000000, 0x00000003,
> >  		0x80032031, 0x1f0c0000, 0xd0061e8c, 0x04000000,
> > @@ -484,13 +485,14 @@ struct iga64_template const iga64_code_media_block_write_aip[] = {
> >  };
> >  
> >  struct iga64_template const iga64_code_common_target_write[] = {
> > -	{ .gen_ver = 2000, .size = 48, .code = (const uint32_t []) {
> > +	{ .gen_ver = 2000, .size = 52, .code = (const uint32_t []) {
> >  		0x80100061, 0x1f054220, 0x00000000, 0x00000000,
> >  		0x80000061, 0x1f054220, 0x00000000, 0xc0ded001,
> >  		0x80000061, 0x1f154220, 0x00000000, 0xc0ded002,
> >  		0x80000061, 0x1f254220, 0x00000000, 0xc0ded003,
> >  		0x80000061, 0x1f354220, 0x00000000, 0xc0ded004,
> >  		0x800c0061, 0x1e054220, 0x00000000, 0x00000000,
> > +		0x80000061, 0x1e554220, 0x00000000, 0x00000000,
> >  		0x80000061, 0x1e654220, 0x00000000, 0xc0ded000,
> >  		0x80000061, 0x1e754220, 0x00000000, 0x0000000f,
> >  		0x80032031, 0x00000000, 0xd00e1e94, 0x04000000,
> > @@ -612,14 +614,15 @@ struct iga64_template const iga64_code_clear_r40[] = {
> >  };
> >  
> >  struct iga64_template const iga64_code_jump_dw_neq[] = {
> > -	{ .gen_ver = 2000, .size = 32, .code = (const uint32_t []) {
> > +	{ .gen_ver = 2000, .size = 36, .code = (const uint32_t []) {
> >  		0x800c0061, 0x1e054220, 0x00000000, 0x00000000,
> > +		0x80000061, 0x1e554220, 0x00000000, 0x00000000,
> >  		0x80000061, 0x1e654220, 0x00000000, 0xc0ded000,
> >  		0x80000061, 0x1e754220, 0x00000000, 0x00000003,
> >  		0x80032031, 0x1f0c0000, 0xd0061e8c, 0x04000000,
> >  		0x80000061, 0x30014220, 0x00000000, 0x00000000,
> >  		0x80008070, 0x00018220, 0x22001f04, 0xc0ded001,
> > -		0x84000020, 0x00004000, 0x00000000, 0xffffffa0,
> > +		0x84000020, 0x00004000, 0x00000000, 0xffffff90,
> >  		0x80000901, 0x00010000, 0x00000000, 0x00000000,
> >  	}},
> >  	{ .gen_ver = 1270, .size = 40, .code = (const uint32_t []) {
> > diff --git a/lib/iga64_macros.h b/lib/iga64_macros.h
> > index 40b6338928e1..f82785610207 100644
> > --- a/lib/iga64_macros.h
> > +++ b/lib/iga64_macros.h
> > @@ -21,6 +21,13 @@
> >  #define R0_TGIDY r0.6<0;1,0>:ud
> >  #define R0_FFTID r0.5<0;1,0>:ud
> >  
> > +/* Inline data from COMPUTE_WALKER*, Bspec: 47203, 73584
> > + * Filled by __xe*_gpgpu_execfunc.
> > + */
> > +#define R1_TGT_ADDRESS r1.0<0;1,0>:uq
> > +#define R1_TGT_WIDTH r1.2<0;1,0>:ud
> > +#define R1_TGT_HEIGHT r1.3<0;1,0>:ud
> > +
> >  #define SET_SHARED_MEDIA_BLOCK_MSG_HDR(dst, y, width)	\
> >  (W)	mov (8)		dst.0<1>:ud	0x0:ud		;\
> >  (W)	mov (1)		dst.1<1>:ud	y		;\
> > @@ -35,28 +42,50 @@
> >  (W)	mov (1)		dst.2<1>:ud	(width - 1):ud		;\
> >  (W)	mov (1)		dst.4<1>:ud	R0_FFTID
> >  
> > +#if GEN_VER < 3000
> > +#define SET_SURFACE_DESC(dst)			\
> > +(W)	mov (8)		dst.0<1>:ud	0x0:ud
> > +#else
> > +#define SET_SURFACE_DESC(dst)					\
> > +(W)	mov (1)		dst.0<1>:uq	R1_TGT_ADDRESS		;\
> > +(W)	add (1)		dst.2<1>:ud	R1_TGT_WIDTH	-1:d	;\
> > +(W)	add (1)		dst.3<1>:ud	R1_TGT_HEIGHT	-1:d	;\
> > +(W)	add (1)		dst.4<1>:ud	R1_TGT_WIDTH	-1:d
> > +#endif
> > +
> >  #define SET_SHARED_MEDIA_A2DBLOCK_PAYLOAD(dst, y, width)	\
> > -(W)	mov (8)		dst.0<1>:ud	0x0:ud		;\
> > -(W)	mov (1)		dst.6<1>:ud	y		;\
> > +	SET_SURFACE_DESC(dst)					;\
> > +(W)	mov (1)		dst.5<1>:ud	0x0:ud			;\
> > +(W)	mov (1)		dst.6<1>:ud	y			;\
> >  (W)	mov (1)		dst.7<1>:ud	(width - 1):ud
> >  
> > -#define SET_THREAD_MEDIA_A2DBLOCK_PAYLOAD(dst, x, y, width)		\
> > -(W)	mov (8)		dst.0<1>:ud	0x0:ud			;\
> > +#define SET_THREAD_MEDIA_A2DBLOCK_PAYLOAD(dst, x, y, width)	\
> > +	SET_SURFACE_DESC(dst)					;\
> >  (W)	shl (1)		dst.5<1>:ud	R0_TGIDX	0x2:ud	;\
> >  (W)	add (1)		dst.5<1>:ud	dst.5<0;1,0>:ud	x:ud	;\
> >  (W)	add (1)		dst.6<1>:ud	R0_TGIDY	y	;\
> >  (W)	mov (1)		dst.7<1>:ud	(width - 1):ud		;\
> >  
> >  #if GEN_VER < 2000
> > +
> >  #define SET_SHARED_SPACE_ADDR(dst, y, width) SET_SHARED_MEDIA_BLOCK_MSG_HDR(dst, y, width)
> >  #define SET_THREAD_SPACE_ADDR(dst, x, y, width) SET_THREAD_MEDIA_BLOCK_MSG_HDR(dst, x, y, width)
> >  #define LOAD_SPACE_DW(dst, src) send.dc1 (1)	dst	src	src1_null 0x0	0x2190000
> >  #define STORE_SPACE_DW(dst, src) send.dc1 (1)	null	dst	null	0x0	0x40A8000
> > +
> >  #else
> > +
> >  #define SET_SHARED_SPACE_ADDR(dst, y, width) SET_SHARED_MEDIA_A2DBLOCK_PAYLOAD(dst, y, width)
> >  #define SET_THREAD_SPACE_ADDR(dst, x, y, width) SET_THREAD_MEDIA_A2DBLOCK_PAYLOAD(dst, x, y, width)
> > +
nit: comparing to other definitions these 4 newlines looks extra. Keep styling same.

Reviewed-by: Dominik Grzegorzek <dominik.grzegorzek@intel.com>

Patch no longer applies. I've tested it by reseting few top commits, so probably this needs resend
anyway.

Regards,
Dominik
> > +#if GEN_VER < 3000
> >  #define LOAD_SPACE_DW(dst, src) send.tgm (1)	dst	src	null:0	0x0	0x62100003
> >  #define STORE_SPACE_DW(dst, src) send.tgm (1)	null	dst	null:0	0x0	0x64000007
> > +#else
> > +#define LOAD_SPACE_DW(dst, src) send.ugm (1)	dst	src	null:0	0x0	0x2120003
> > +#define STORE_SPACE_DW(dst, src) send.ugm (1)	null	dst	src:1	0x0	0x2020007
> > +#endif
> > +
> >  #endif
> >  
> >  #endif
> > 


  reply	other threads:[~2024-11-25 14:08 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-11-25  7:31 [PATCH v4 0/4] lib/gpgpu_shader: simplify load/store shaders and add Xe3 support Andrzej Hajda
2024-11-25  7:31 ` [PATCH v4 1/4] scripts/generate_iga64_codes: add iga64_macros.h to checksum calculation Andrzej Hajda
2024-11-25  7:31 ` [PATCH v4 2/4] lib/gpgpu_shader: simplify load/store shaders Andrzej Hajda
2024-11-25  7:31 ` [PATCH v4 3/4] lib/gpgpu_shader: pass surface desription to shaders via inline data Andrzej Hajda
2024-11-25  8:50   ` Grzegorzek, Dominik
2024-11-25  7:31 ` [PATCH v4 4/4] lib/gpgpu_shader: add support for Xe3 platforms Andrzej Hajda
2024-11-25 14:08   ` Grzegorzek, Dominik [this message]
2024-11-25  9:19 ` ✓ Xe.CI.BAT: success for lib/gpgpu_shader: simplify load/store shaders and add Xe3 support (rev2) Patchwork
2024-11-25  9:29 ` ✗ i915.CI.BAT: failure " Patchwork
2024-11-25 10:40 ` ✗ Xe.CI.Full: " Patchwork

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=7f1698d519d08d158043339681300cee95da2b0e.camel@intel.com \
    --to=dominik.grzegorzek@intel.com \
    --cc=andrzej.hajda@intel.com \
    --cc=christoph.manszewski@intel.com \
    --cc=gwan-gyeong.mun@intel.com \
    --cc=igt-dev@lists.freedesktop.org \
    --cc=kamil.konieczny@linux.intel.com \
    --cc=zbigniew.kempczynski@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox