AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

panfrost/va: valhall-specific swizzle lowering

Browse Source 
For most swizzled instructions that are different between valhall and
bifrost, valhall allows more values than bifrost does, so we can avoid
some unnecessary lowering.

Signed-off-by: Benjamin Lee <benjamin.lee@collabora.com>
Reviewed-by: Eric R. Smith <eric.smith@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33416>
This commit is contained in:
Benjamin Lee
2025-02-13 03:03:02 -08:00
parent 168c96816a
commit b683a59438
1 changed files with 81 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/panfrost/compiler/bi_lower_swizzle.c
+81 -1
View File
@@ -25,6 +25,7 @@
#include "bi_swizzles.h"
#include "compiler.h"
#include "valhall.h"
#include "panfrost/lib/pan_props.h"
/* Not all 8-bit and 16-bit instructions support all swizzles on all sources.
* These passes, intended to run after NIR->BIR but before scheduling/RA, lower
@@ -46,6 +47,84 @@ bi_swizzle_replicates_8(enum bi_swizzle swz)
}
}
static uint32_t
va_op_swizzles(enum bi_opcode op, unsigned src)
{
/* This is a bifrost-only instruction that is lowered on valhall */
if (!valhall_opcodes[op].exact)
return bi_op_swizzles[op][src];
uint32_t swizzles = 0;
struct va_src_info info = va_src_info(op, src);
if (info.swizzle) {
assert(info.size == VA_SIZE_16 || info.size == VA_SIZE_32);
if (info.size == VA_SIZE_16)
swizzles |= (1 << BI_SWIZZLE_H00) | (1 << BI_SWIZZLE_H10) |
(1 << BI_SWIZZLE_H01) | (1 << BI_SWIZZLE_H11);
else if (info.size == VA_SIZE_32)
swizzles |= (1 << BI_SWIZZLE_H01) | (1 << BI_SWIZZLE_H0) |
(1 << BI_SWIZZLE_H1);
}
if (info.lane) {
if (info.size == VA_SIZE_8)
swizzles |= (1 << BI_SWIZZLE_B0) | (1 << BI_SWIZZLE_B1) |
(1 << BI_SWIZZLE_B2) | (1 << BI_SWIZZLE_B3);
if (info.size == VA_SIZE_16)
swizzles |= (1 << BI_SWIZZLE_H0) | (1 << BI_SWIZZLE_H1);
}
if (info.lanes) {
assert(info.size == VA_SIZE_8);
swizzles |= (1 << BI_SWIZZLE_B00) | (1 << BI_SWIZZLE_B11) |
(1 << BI_SWIZZLE_B22) | (1 << BI_SWIZZLE_B33);
}
if (info.halfswizzle) {
assert(info.size == VA_SIZE_8);
swizzles |= (1 << BI_SWIZZLE_B00) | (1 << BI_SWIZZLE_B10) |
(1 << BI_SWIZZLE_B20) | (1 << BI_SWIZZLE_B30) |
(1 << BI_SWIZZLE_B01) | (1 << BI_SWIZZLE_B11) |
(1 << BI_SWIZZLE_B21) | (1 << BI_SWIZZLE_B31) |
(1 << BI_SWIZZLE_B02) | (1 << BI_SWIZZLE_B12) |
(1 << BI_SWIZZLE_B22) | (1 << BI_SWIZZLE_B32) |
(1 << BI_SWIZZLE_B03) | (1 << BI_SWIZZLE_B13) |
(1 << BI_SWIZZLE_B23) | (1 << BI_SWIZZLE_B33);
}
if (info.widen) {
if (info.size == VA_SIZE_8)
swizzles |= (1 << BI_SWIZZLE_B0123) | (1 << BI_SWIZZLE_B0101) |
(1 << BI_SWIZZLE_B2323) | (1 << BI_SWIZZLE_B0000) |
(1 << BI_SWIZZLE_B1111) | (1 << BI_SWIZZLE_B2222) |
(1 << BI_SWIZZLE_B3333);
else if (info.size == VA_SIZE_16)
swizzles |= (1 << BI_SWIZZLE_H00) | (1 << BI_SWIZZLE_H10) |
(1 << BI_SWIZZLE_H01) | (1 << BI_SWIZZLE_H11) |
(1 << BI_SWIZZLE_B00) | (1 << BI_SWIZZLE_B11) |
(1 << BI_SWIZZLE_B22) | (1 << BI_SWIZZLE_B33) |
(1 << BI_SWIZZLE_B01) | (1 << BI_SWIZZLE_B20) |
(1 << BI_SWIZZLE_B02) | (1 << BI_SWIZZLE_B31) |
(1 << BI_SWIZZLE_B13) | (1 << BI_SWIZZLE_B23);
else if (info.size == VA_SIZE_32)
swizzles |= (1 << BI_SWIZZLE_H01) | (1 << BI_SWIZZLE_H0) |
(1 << BI_SWIZZLE_H1) | (1 << BI_SWIZZLE_B0) |
(1 << BI_SWIZZLE_B1) | (1 << BI_SWIZZLE_B2) |
(1 << BI_SWIZZLE_B3);
}
if (info.combine) {
assert(info.size == VA_SIZE_32);
swizzles |= (1 << BI_SWIZZLE_H01) | (1 << BI_SWIZZLE_H0) | (1 << BI_SWIZZLE_H1);
}
if (swizzles == 0)
swizzles = 1 << BI_SWIZZLE_H01;
return swizzles;
}
static void
lower_swizzle(bi_context *ctx, bi_instr *ins, unsigned src)
{
@@ -63,7 +142,8 @@ lower_swizzle(bi_context *ctx, bi_instr *ins, unsigned src)
return;
}
uint32_t supported_swizzles = bi_op_swizzles[ins->op][src];
uint32_t supported_swizzles = pan_arch(ctx->inputs->gpu_id) >= 9 ?
va_op_swizzles(ins->op, src) : bi_op_swizzles[ins->op][src];
if (supported_swizzles & (1 << ins->src[src].swizzle))
return;