Mesa (master): ir3: Assemble and disassemble swz/gat/sct

GitLab Mirror gitlab-mirror at kemper.freedesktop.org
Mon Apr 19 16:32:40 UTC 2021


Module: Mesa
Branch: master
Commit: 08499369d0164e9a503e16aad39f9b9e1e347e1a
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=08499369d0164e9a503e16aad39f9b9e1e347e1a

Author: Connor Abbott <cwabbott0 at gmail.com>
Date:   Fri Apr 16 17:13:30 2021 +0200

ir3: Assemble and disassemble swz/gat/sct

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10291>

---

 src/freedreno/ir3/instr-a3xx.h   |   9 ++
 src/freedreno/ir3/ir3_lexer.l    |   3 +
 src/freedreno/ir3/ir3_parser.y   |  12 +++
 src/freedreno/ir3/tests/disasm.c |   5 +
 src/freedreno/isa/ir3-cat1.xml   | 192 +++++++++++++++++++++++++++++++++------
 5 files changed, 193 insertions(+), 28 deletions(-)

diff --git a/src/freedreno/ir3/instr-a3xx.h b/src/freedreno/ir3/instr-a3xx.h
index b56983a3b2b..b7f2f95ccda 100644
--- a/src/freedreno/ir3/instr-a3xx.h
+++ b/src/freedreno/ir3/instr-a3xx.h
@@ -92,8 +92,17 @@ typedef enum {
 
 	/* category 1: */
 	OPC_MOV             = _OPC(1, 0),
+	OPC_MOVP            = _OPC(1, 1),
+	/* swz, gat, sct */
 	OPC_MOVMSK          = _OPC(1, 3),
 
+	/* Virtual opcodes for instructions differentiated via a "sub-opcode" that
+	 * replaces the repeat field:
+	 */
+	OPC_SWZ            = _OPC(1, 4),
+	OPC_GAT            = _OPC(1, 5),
+	OPC_SCT            = _OPC(1, 6),
+
 	/* Logical opcodes for different variants of mov: */
 	OPC_MOV_IMMED       = _OPC(1, 40),
 	OPC_MOV_CONST       = _OPC(1, 41),
diff --git a/src/freedreno/ir3/ir3_lexer.l b/src/freedreno/ir3/ir3_lexer.l
index 568145af71e..f7d9783c628 100644
--- a/src/freedreno/ir3/ir3_lexer.l
+++ b/src/freedreno/ir3/ir3_lexer.l
@@ -180,6 +180,9 @@ static int parse_w(const char *str)
 "mova"                            return TOKEN(T_OP_MOVA);
 "mov"                             return TOKEN(T_OP_MOV);
 "cov"                             return TOKEN(T_OP_COV);
+"swz"                             return TOKEN(T_OP_SWZ);
+"gat"                             return TOKEN(T_OP_GAT);
+"sct"                             return TOKEN(T_OP_SCT);
 
 ("f16"|"f32"|"u16"|"u32"|"s16"|"s32"|"u8"|"s8"){2} ir3_yylval.str = yytext; return T_CAT1_TYPE_TYPE;
 
diff --git a/src/freedreno/ir3/ir3_parser.y b/src/freedreno/ir3/ir3_parser.y
index fc684a54159..ae23c6a1da0 100644
--- a/src/freedreno/ir3/ir3_parser.y
+++ b/src/freedreno/ir3/ir3_parser.y
@@ -399,6 +399,9 @@ static void print_token(FILE *file, int type, YYSTYPE value)
 %token <tok> T_OP_MOVA
 %token <tok> T_OP_MOV
 %token <tok> T_OP_COV
+%token <tok> T_OP_SWZ
+%token <tok> T_OP_GAT
+%token <tok> T_OP_SCT
 
 /* category 2: */
 %token <tok> T_OP_ADD_F
@@ -791,10 +794,19 @@ cat1_mova:         T_OP_MOVA T_A0 ',' {
                        new_reg((61 << 3), IR3_REG_HALF);
                    } cat1_src
 
+cat1_swz:          T_OP_SWZ '.' T_CAT1_TYPE_TYPE { parse_type_type(new_instr(OPC_SWZ), $3); } dst_reg ',' dst_reg ',' src_reg ',' src_reg
+
+cat1_gat:          T_OP_GAT '.' T_CAT1_TYPE_TYPE { parse_type_type(new_instr(OPC_GAT), $3); } dst_reg ',' src_reg ',' src_reg ',' src_reg ',' src_reg
+
+cat1_sct:          T_OP_SCT '.' T_CAT1_TYPE_TYPE { parse_type_type(new_instr(OPC_SCT), $3); } dst_reg ',' dst_reg ',' dst_reg ',' dst_reg ',' src_reg
+
                    /* NOTE: cat1 can also *write* to relative gpr */
 cat1_instr:        cat1_movmsk
 |                  cat1_mova1
 |                  cat1_mova
+|                  cat1_swz
+|                  cat1_gat
+|                  cat1_sct
 |                  cat1_opc dst_reg ',' cat1_src
 |                  cat1_opc relative_gpr ',' cat1_src
 
diff --git a/src/freedreno/ir3/tests/disasm.c b/src/freedreno/ir3/tests/disasm.c
index 5d6d052f7fb..6f500f0253e 100644
--- a/src/freedreno/ir3/tests/disasm.c
+++ b/src/freedreno/ir3/tests/disasm.c
@@ -84,6 +84,11 @@ static const struct test {
 	/* dEQP-VK.subgroups.ballot.compute.compute */
 	INSTR_6XX(260cc3c0_00000000, "movmsk.w128 r48.x"), /* movmsk.w128 sr48.x */
 
+	INSTR_6XX(240cc004_00030201, "swz.u32u32 r1.x, r0.w, r0.y, r0.z"),
+	INSTR_6XX(2400c105_04030201, "gat.f16u32 r1.y, hr0.y, hr0.z, hr0.w, hr1.x"),
+	INSTR_6XX(240c0205_04030201, "sct.u32f16 hr1.y, hr0.z, hr0.w, hr1.x, r0.y"),
+	INSTR_6XX(2400c205_04030201, "sct.f16u32 r1.y, r0.z, r0.w, r1.x, hr0.y"),
+
 	INSTR_6XX(20510005_0000ffff, "mov.s16s16 hr1.y, -1"),
 	INSTR_6XX(20400005_00003900, "mov.f16f16 hr1.y, h(0.625000)"),
 	INSTR_6XX(20400006_00003800, "mov.f16f16 hr1.z, h(0.500000)"),
diff --git a/src/freedreno/isa/ir3-cat1.xml b/src/freedreno/isa/ir3-cat1.xml
index 671ccd42e2b..2633894a05e 100644
--- a/src/freedreno/isa/ir3-cat1.xml
+++ b/src/freedreno/isa/ir3-cat1.xml
@@ -68,14 +68,9 @@ SOFTWARE.
 </enum>
 
 <bitset name="#instruction-cat1" extends="#instruction">
-	<field name="DST" low="32" high="39" type="#cat1-dst">
-		<param name="DST_REL"/>
-	</field>
-	<field name="REPEAT" low="40" high="41" type="#rptN"/>
 	<pattern pos="42">0</pattern>
 	<field name="SS" pos="44" type="bool" display="(ss)"/>
 	<field name="UL" pos="45" type="bool" display="(ul)"/>
-	<field name="DST_REL" pos="49" type="bool"/>
 	<field name="ROUND" low="55" high="56" type="#round"/>
 	<field name="JP" pos="59" type="bool" display="(jp)"/>
 	<field name="SY" pos="60" type="bool" display="(sy)"/>
@@ -84,14 +79,40 @@ SOFTWARE.
 		<map name="SRC">src->regs[1]</map>
 		<map name="SRC_R">!!(src->regs[1]->flags & IR3_REG_R)</map>
 		<map name="UL">!!(src->flags & IR3_INSTR_UL)</map>
-		<map name="DST_TYPE">src->cat1.dst_type</map>
 		<map name="DST_REL">!!(src->regs[0]->flags & IR3_REG_RELATIV)</map>
-		<map name="SRC_TYPE">src->cat1.src_type</map>
 		<map name="ROUND">src->cat1.round</map>
 	</encode>
 </bitset>
 
-<bitset name="#instruction-cat1-mov" extends="#instruction-cat1">
+<bitset name="#instruction-cat1-typed" extends="#instruction-cat1">
+	<derived name="HALF" type="bool" display="h">
+		<expr>
+			({SRC_TYPE} == 0) /* f16 */ ||
+			({SRC_TYPE} == 2) /* u16 */ ||
+			({SRC_TYPE} == 4) /* s16 */ ||
+			({SRC_TYPE} == 6) /* u8 */  ||
+			({SRC_TYPE} == 7) /* s8 */
+		</expr>
+	</derived>
+	<derived name="DST_HALF" type="bool" display="h">
+		<expr>
+			({DST_TYPE} == 0) /* f16 */ ||
+			({DST_TYPE} == 2) /* u16 */ ||
+			({DST_TYPE} == 4) /* s16 */ ||
+			({DST_TYPE} == 6) /* u8 */  ||
+			({DST_TYPE} == 7) /* s8 */
+		</expr>
+	</derived>
+	<field name="DST_TYPE" low="46" high="48" type="#type"/>
+	<field name="SRC_TYPE" low="50" high="52" type="#type"/>
+
+	<encode>
+		<map name="DST_TYPE">src->cat1.dst_type</map>
+		<map name="SRC_TYPE">src->cat1.src_type</map>
+	</encode>
+</bitset>
+
+<bitset name="#instruction-cat1-mov" extends="#instruction-cat1-typed">
 	<override>
 		<expr>
 			({DST} == 0xf4 /* a0.x */) && ({SRC_TYPE} == 4 /* s16 */) && ({DST_TYPE} == 4)
@@ -125,27 +146,12 @@ SOFTWARE.
 	<display>
 		{SY}{SS}{JP}{REPEAT}{UL}mov.{SRC_TYPE}{DST_TYPE} {ROUND}{DST_HALF}{DST}, {SRC}
 	</display>
+	<field name="DST" low="32" high="39" type="#cat1-dst">
+		<param name="DST_REL"/>
+	</field>
+	<field name="REPEAT" low="40" high="41" type="#rptN"/>
+	<field name="DST_REL" pos="49" type="bool"/>
 	<pattern low="57" high="58">00</pattern>  <!-- OPC -->
-	<derived name="HALF" type="bool" display="h">
-		<expr>
-			({SRC_TYPE} == 0) /* f16 */ ||
-			({SRC_TYPE} == 2) /* u16 */ ||
-			({SRC_TYPE} == 4) /* s16 */ ||
-			({SRC_TYPE} == 6) /* u8 */  ||
-			({SRC_TYPE} == 7) /* s8 */
-		</expr>
-	</derived>
-	<derived name="DST_HALF" type="bool" display="h">
-		<expr>
-			({DST_TYPE} == 0) /* f16 */ ||
-			({DST_TYPE} == 2) /* u16 */ ||
-			({DST_TYPE} == 4) /* s16 */ ||
-			({DST_TYPE} == 6) /* u8 */  ||
-			({DST_TYPE} == 7) /* s8 */
-		</expr>
-	</derived>
-	<field name="DST_TYPE" low="46" high="48" type="#type"/>
-	<field name="SRC_TYPE" low="50" high="52" type="#type"/>
 </bitset>
 
 <!--
@@ -304,6 +310,131 @@ SOFTWARE.
 	Other newer cat1 instructions
  -->
 
+<bitset name="#cat1-multi-src" size="8">
+	<display>
+		{HALF}{REG}
+	</display>
+	<field name="REG" low="0" high="7" type="#reg-gpr"/>
+
+	<encode type="struct ir3_register *">
+		<map name="REG">src</map>
+	</encode>
+</bitset>
+
+<bitset name="#cat1-multi-dst" size="8">
+	<display>
+		{DST_HALF}{REG}
+	</display>
+
+	<field name="REG" low="0" high="7" type="#reg-gpr"/>
+
+	<encode type="struct ir3_register *">
+		<map name="REG">src</map>
+	</encode>
+</bitset>
+
+<bitset name="#instruction-cat1-multi" extends="#instruction-cat1-typed">
+	<doc>
+		These instructions all expand to a series of mov instructions,
+		like (rptN) but more flexible. They aren't any faster than the
+		equivalent sequence of mov/cov, but they guarantee that all
+		sources are read before any destination is written, so they
+		behave as-if the moves are executed in parallel.
+		TODO: when were these added?
+	</doc>
+	<field name="SRC0" low="0" high="7" type="#cat1-multi-src">
+		<param name="HALF"/>
+	</field>
+	<field name="DST0" low="32" high="39" type="#cat1-multi-dst">
+		<param name="DST_HALF"/>
+	</field>
+	<pattern pos="43">0</pattern>   <!-- SRC_R -->
+	<pattern pos="49">0</pattern>   <!-- DST_REL -->
+	<pattern low="53" high="54">00</pattern>
+	<pattern low="57" high="58">10</pattern>  <!-- OPC -->
+
+	<encode>
+		<map name="DST0">src->regs[0]</map>
+	</encode>
+</bitset>
+
+<bitset name="swz" extends="#instruction-cat1-multi">
+	<doc>
+                SWiZzle. Move SRC0 to DST0 and SRC1 to DST1 in parallel. In
+                particular this can be used to swap two registers.
+	</doc>
+	<display>
+		{SY}{SS}{JP}{UL}swz.{SRC_TYPE}{DST_TYPE} {ROUND}{DST0}, {DST1}, {SRC0}, {SRC1}
+	</display>
+	<field name="SRC1" low="8" high="15" type="#cat1-multi-src">
+		<param name="HALF"/>
+	</field>
+	<field name="DST1" low="16" high="23" type="#cat1-multi-dst">
+		<param name="DST_HALF"/>
+	</field>
+	<pattern low="24" high="31">00000000</pattern>
+
+	<pattern low="40" high="41">00</pattern> <!-- SUB_OPC -->
+
+	<encode>
+		<map name="SRC0">src->regs[2]</map>
+		<map name="SRC1">src->regs[3]</map>
+		<map name="DST1">src->regs[1]</map>
+	</encode>
+</bitset>
+
+<bitset name="gat" extends="#instruction-cat1-multi">
+	<doc>
+		GATher. Move SRC0 to DST0, SRC1 to DST0 + 1, SRC2 to DST0 + 2, and SRC3 to DST0 + 3.
+	</doc>
+	<display>
+		{SY}{SS}{JP}{UL}gat.{SRC_TYPE}{DST_TYPE} {ROUND}{DST0}, {SRC0}, {SRC1}, {SRC2}, {SRC3}
+	</display>
+	<field name="SRC1" low="8" high="15" type="#cat1-multi-src">
+		<param name="HALF"/>
+	</field>
+	<field name="SRC2" low="16" high="23" type="#cat1-multi-src">
+		<param name="HALF"/>
+	</field>
+	<field name="SRC3" low="24" high="31" type="#cat1-multi-src">
+		<param name="HALF"/>
+	</field>
+	<pattern low="40" high="41">01</pattern> <!-- SUB_OPC -->
+
+	<encode>
+		<map name="SRC0">src->regs[1]</map>
+		<map name="SRC1">src->regs[2]</map>
+		<map name="SRC2">src->regs[3]</map>
+		<map name="SRC3">src->regs[4]</map>
+	</encode>
+</bitset>
+
+<bitset name="sct" extends="#instruction-cat1-multi">
+	<doc>
+		SCaTter. Move SRC0 to DST0, SRC0 + 1 to DST1, SRC0 + 2 to DST2 + 3, and SRC0 + 3 to DST3.
+	</doc>
+	<display>
+		{SY}{SS}{JP}{UL}sct.{SRC_TYPE}{DST_TYPE} {ROUND}{DST0}, {DST1}, {DST2}, {DST3}, {SRC0}
+	</display>
+	<field name="DST1" low="8" high="15" type="#cat1-multi-dst">
+		<param name="DST_HALF"/>
+	</field>
+	<field name="DST2" low="16" high="23" type="#cat1-multi-dst">
+		<param name="DST_HALF"/>
+	</field>
+	<field name="DST3" low="24" high="31" type="#cat1-multi-dst">
+		<param name="DST_HALF"/>
+	</field>
+	<pattern low="40" high="41">10</pattern> <!-- SUB_OPC -->
+
+	<encode>
+		<map name="SRC0">src->regs[4]</map>
+		<map name="DST1">src->regs[1]</map>
+		<map name="DST2">src->regs[2]</map>
+		<map name="DST3">src->regs[3]</map>
+	</encode>
+</bitset>
+
 <bitset name="movmsk" extends="#instruction-cat1">
 	<display>
 		{SY}{SS}{JP}{UL}movmsk.w{W} {DST}
@@ -314,8 +445,13 @@ SOFTWARE.
 		</expr>
 	</derived>
 	<pattern low="0" high="31">00000000000000000000000000000000</pattern>
+	<field name="DST" low="32" high="39" type="#cat1-dst">
+		<param name="DST_REL"/>
+	</field>
+	<field name="REPEAT" low="40" high="41" type="#rptN"/>
 	<pattern pos="43">0</pattern>   <!-- SRC_R -->
 	<pattern low="46" high="48">011</pattern>       <!-- DST_TYPE==u32 -->
+	<field name="DST_REL" pos="49" type="bool"/>
 	<pattern low="50" high="52">011</pattern>       <!-- SRC_TYPE==u32 -->
 	<pattern low="53" high="54">00</pattern>
 	<pattern low="57" high="58">11</pattern>  <!-- OPC -->



More information about the mesa-commit mailing list