From 90537d6a891df2cefa99b0fdcfec19dbeb8c5a78 Mon Sep 17 00:00:00 2001
From: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Date: Tue, 20 Dec 2016 00:11:33 +0100
Subject: [PATCH] nv50/ir: use sched control codes for gm107 builtins

Yes, IMUL/IMAD require dependency barriers and we should
definitely replace these instructions by XMAD but the
different flags need to be figured out. Note that XMAD only
supports 16-bits integers.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>
---
 .../drivers/nouveau/codegen/lib/gm107.asm     | 40 +++++++++----------
 .../drivers/nouveau/codegen/lib/gm107.asm.h   | 40 +++++++++----------
 2 files changed, 40 insertions(+), 40 deletions(-)

diff --git a/src/gallium/drivers/nouveau/codegen/lib/gm107.asm b/src/gallium/drivers/nouveau/codegen/lib/gm107.asm
index 67b98dac39c..90741b6c59f 100644
--- a/src/gallium/drivers/nouveau/codegen/lib/gm107.asm
+++ b/src/gallium/drivers/nouveau/codegen/lib/gm107.asm
@@ -11,39 +11,39 @@
 // SIZE:    22 / 14 * 8 bytes
 //
 gm107_div_u32:
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0xd wr 0x0 wt 0x3f) (st 0x1 wt 0x1) (st 0x6)
    flo u32 $r2 $r1
    lop xor 1 $r2 $r2 0x1f
    mov $r3 0x1 0xf
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x1) (st 0xf wr 0x0) (st 0x6 wr 0x0 wt 0x1)
    shl $r2 $r3 $r2
    i2i u32 u32 $r1 neg $r1
    imul u32 u32 $r3 $r1 $r2
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x6 wr 0x0 wt 0x1) (st 0x6 wr 0x0 wt 0x1) (st 0x6 wr 0x0 wt 0x1)
    imad u32 u32 hi $r2 $r2 $r3 $r2
    imul u32 u32 $r3 $r1 $r2
    imad u32 u32 hi $r2 $r2 $r3 $r2
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x6 wr 0x0 wt 0x1) (st 0x6 wr 0x0 wt 0x1) (st 0x6 wr 0x0 wt 0x1)
    imul u32 u32 $r3 $r1 $r2
    imad u32 u32 hi $r2 $r2 $r3 $r2
    imul u32 u32 $r3 $r1 $r2
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x6 wr 0x0 wt 0x1) (st 0x6 wr 0x0 wt 0x1) (st 0x6 wr 0x0 wt 0x1)
    imad u32 u32 hi $r2 $r2 $r3 $r2
    imul u32 u32 $r3 $r1 $r2
    imad u32 u32 hi $r2 $r2 $r3 $r2
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x6) (st 0x6 wr 0x0 rd 0x1 wt 0x1) (st 0xf wr 0x0 rd 0x1 wt 0x2)
    mov $r3 $r0 0xf
    imul u32 u32 hi $r0 $r0 $r2
    i2i u32 u32 $r2 neg $r1
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x6 wr 0x0 wt 0x3) (st 0xd wt 0x1) (st 0x1)
    imad u32 u32 $r1 $r1 $r0 $r3
    isetp ge u32 and $p0 1 $r1 $r2 1
    $p0 iadd $r1 $r1 neg $r2
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x5) (st 0xd) (st 0x1)
    $p0 iadd $r0 $r0 0x1
    $p0 isetp ge u32 and $p0 1 $r1 $r2 1
    $p0 iadd $r1 $r1 neg $r2
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x1) (st 0xf) (st 0xf)
    $p0 iadd $r0 $r0 0x1
    ret
    nop 0
@@ -55,47 +55,47 @@ gm107_div_u32:
 // CLOBBER: $r2 - $r3, $p0 - $p3
 //
 gm107_div_s32:
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0xd wt 0x3f) (st 0x1) (st 0x1 wr 0x0)
    isetp lt and $p2 0x1 $r0 0 1
    isetp lt xor $p3 1 $r1 0 $p2
    i2i s32 s32 $r0 abs $r0
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0xf wr 0x1) (st 0xd wr 0x1 wt 0x2) (st 0x1 wt 0x2)
    i2i s32 s32 $r1 abs $r1
    flo u32 $r2 $r1
    lop xor 1 $r2 $r2 0x1f
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x6) (st 0x1) (st 0xf wr 0x1)
    mov $r3 0x1 0xf
    shl $r2 $r3 $r2
    i2i u32 u32 $r1 neg $r1
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x6 wr 0x1 wt 0x2) (st 0x6 wr 0x1 wt 0x2) (st 0x6 wr 0x1 wt 0x2)
    imul u32 u32 $r3 $r1 $r2
    imad u32 u32 hi $r2 $r2 $r3 $r2
    imul u32 u32 $r3 $r1 $r2
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x6 wr 0x1 wt 0x2) (st 0x6 wr 0x1 wt 0x2) (st 0x6 wr 0x1 wt 0x2)
    imad u32 u32 hi $r2 $r2 $r3 $r2
    imul u32 u32 $r3 $r1 $r2
    imad u32 u32 hi $r2 $r2 $r3 $r2
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x6 wr 0x1 wt 0x2) (st 0x6 wr 0x1 wt 0x2) (st 0x6 wr 0x1 wt 0x2)
    imul u32 u32 $r3 $r1 $r2
    imad u32 u32 hi $r2 $r2 $r3 $r2
    imul u32 u32 $r3 $r1 $r2
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x6 wr 0x1 rd 0x2 wt 0x2) (st 0x2 wt 0x5) (st 0x6 wr 0x0 rd 0x1 wt 0x2)
    imad u32 u32 hi $r2 $r2 $r3 $r2
    mov $r3 $r0 0xf
    imul u32 u32 hi $r0 $r0 $r2
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0xf wr 0x1 rd 0x2 wt 0x2) (st 0x6 wr 0x0 wt 0x5) (st 0xd wt 0x3)
    i2i u32 u32 $r2 neg $r1
    imad u32 u32 $r1 $r1 $r0 $r3
    isetp ge u32 and $p0 1 $r1 $r2 1
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x1) (st 0x5) (st 0xd)
    $p0 iadd $r1 $r1 neg $r2
    $p0 iadd $r0 $r0 0x1
    $p0 isetp ge u32 and $p0 1 $r1 $r2 1
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0x1) (st 0x2) (st 0xf wr 0x0)
    $p0 iadd $r1 $r1 neg $r2
    $p0 iadd $r0 $r0 0x1
    $p3 i2i s32 s32 $r0 neg $r0
-   sched (st 0x0) (st 0x0) (st 0x0)
+   sched (st 0xf wr 0x1) (st 0xf wt 0x3) (st 0xf)
    $p2 i2i s32 s32 $r1 neg $r1
    ret
    nop 0
diff --git a/src/gallium/drivers/nouveau/codegen/lib/gm107.asm.h b/src/gallium/drivers/nouveau/codegen/lib/gm107.asm.h
index 7be25da5532..8708a94b0a6 100644
--- a/src/gallium/drivers/nouveau/codegen/lib/gm107.asm.h
+++ b/src/gallium/drivers/nouveau/codegen/lib/gm107.asm.h
@@ -1,83 +1,83 @@
 uint64_t gm107_builtin_code[] = {
 /* 0x0000: gm107_div_u32 */
-	0x001f8000fc0007e0,
+	0x001f9801fc21ff0d,
 	0x5c30000000170002,
 	0x3847040001f70202,
 	0x3898078000170003,
-	0x001f8000fc0007e0,
+	0x003c1800e1e007e1,
 	0x5c48000000270302,
 	0x5ce0200000170a01,
 	0x5c38000000270103,
-	0x001f8000fc0007e0,
+	0x003c1801e0c00f06,
 	0x5a40010000370202,
 	0x5c38000000270103,
 	0x5a40010000370202,
-	0x001f8000fc0007e0,
+	0x003c1801e0c00f06,
 	0x5c38000000270103,
 	0x5a40010000370202,
 	0x5c38000000270103,
-	0x001f8000fc0007e0,
+	0x003c1801e0c00f06,
 	0x5a40010000370202,
 	0x5c38000000270103,
 	0x5a40010000370202,
-	0x001f8000fc0007e0,
+	0x00443c0120c007e6,
 	0x5c98078000070003,
 	0x5c38008000270000,
 	0x5ce0200000170a02,
-	0x001f8000fc0007e0,
+	0x001f8401fda01f06,
 	0x5a00018000070101,
 	0x5b6c038000270107,
 	0x5c11000000200101,
-	0x001f8000fc0007e0,
+	0x001f8400fda007e5,
 	0x3810000000100000,
 	0x5b6c038000200107,
 	0x5c11000000200101,
-	0x001f8000fc0007e0,
+	0x001fbc00fde007e1,
 	0x3810000000100000,
 	0xe32000000007000f,
 	0x50b0000000070f00,
 /* 0x0120: gm107_div_s32 */
-	0x001f8000fc0007e0,
+	0x001c0400fc21ffed,
 	0x5b6303800ff70017,
 	0x5b6341000ff7011f,
 	0x5ce2000000073a00,
-	0x001f8000fc0007e0,
+	0x005f8402e5a0072f,
 	0x5ce2000000173a01,
 	0x5c30000000170002,
 	0x3847040001f70202,
-	0x001f8000fc0007e0,
+	0x001cbc00fc2007e6,
 	0x3898078000170003,
 	0x5c48000000270302,
 	0x5ce0200000170a01,
-	0x001f8000fc0007e0,
+	0x005c9802e4c01726,
 	0x5c38000000270103,
 	0x5a40010000370202,
 	0x5c38000000270103,
-	0x001f8000fc0007e0,
+	0x005c9802e4c01726,
 	0x5a40010000370202,
 	0x5c38000000270103,
 	0x5a40010000370202,
-	0x001f8000fc0007e0,
+	0x005c9802e4c01726,
 	0x5c38000000270103,
 	0x5a40010000370202,
 	0x5c38000000270103,
-	0x001f8000fc0007e0,
+	0x00441805fc401226,
 	0x5a40010000370202,
 	0x5c98078000070003,
 	0x5c38008000270000,
-	0x001f8000fc0007e0,
+	0x007fb405e0c0122f,
 	0x5ce0200000170a02,
 	0x5a00018000070101,
 	0x5b6c038000270107,
-	0x001f8000fc0007e0,
+	0x001fb400fca007e1,
 	0x5c11000000200101,
 	0x3810000000100000,
 	0x5b6c038000200107,
-	0x001f8000fc0007e0,
+	0x001c3c00fc4007e1,
 	0x5c11000000200101,
 	0x3810000000100000,
 	0x5ce0200000033a00,
-	0x001f8000fc0007e0,
+	0x001fbc03fde0072f,
 	0x5ce0200000123a01,
 	0xe32000000007000f,
 	0x50b0000000070f00,
-- 
2.30.2