diff options
| author | Samuel Pitoiset <[email protected]> | 2015-10-09 11:22:20 +0200 |
|---|---|---|
| committer | Samuel Pitoiset <[email protected]> | 2015-10-16 21:57:44 +0200 |
| commit | c4896c99cbe10b829981250465baf0b00e18ba40 (patch) | |
| tree | 7d434918a651d126688dc266d928a095632466ce /src/gallium/drivers/nouveau/nvc0 | |
| parent | 7abd707251f29aaf27f83644e47d2dc8b75e10c6 (diff) | |
nvc0: fix unaligned mem access when reading MP counters on Fermi
Memory access have to be aligned to 128-bits. Note that this
doesn't happen when the card only has TPC.
This patch fixes the following dmesg fail:
gr: GPC0/TPC1/MP trap: global 00000004 [MULTIPLE_WARP_ERRORS] warp 000f
[UNALIGNED_MEM_ACCESS]
Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Ilia Mirkin <[email protected]>
Diffstat (limited to 'src/gallium/drivers/nouveau/nvc0')
| -rw-r--r-- | src/gallium/drivers/nouveau/nvc0/nvc0_query_hw_sm.c | 18 |
1 files changed, 12 insertions, 6 deletions
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_query_hw_sm.c b/src/gallium/drivers/nouveau/nvc0/nvc0_query_hw_sm.c index 2f827b0a944..8eb3b3ef14a 100644 --- a/src/gallium/drivers/nouveau/nvc0/nvc0_query_hw_sm.c +++ b/src/gallium/drivers/nouveau/nvc0/nvc0_query_hw_sm.c @@ -247,7 +247,7 @@ static const uint64_t nvc0_read_hw_sm_counters_code[] = * mov b32 $r11 c0[0x4] * ext u32 $r8 $r9 0x414 * (not $p0) exit - * mul $r8 u32 $r8 u32 36 + * mul $r8 u32 $r8 u32 48 * add b32 $r10 $c $r10 $r8 * add b32 $r11 $r11 0x0 $c * mov b32 $r8 c0[0x8] @@ -270,7 +270,7 @@ static const uint64_t nvc0_read_hw_sm_counters_code[] = 0x280040001002dde4ULL, 0x7000c01050921c03ULL, 0x80000000000021e7ULL, - 0x1000000090821c02ULL, + 0x10000000c0821c02ULL, 0x4801000020a29c03ULL, 0x0800000000b2dc42ULL, 0x2800400020021de4ULL, @@ -473,7 +473,7 @@ nvc0_hw_sm_begin_query(struct nvc0_context *nvc0, struct nvc0_hw_query *hq) /* set sequence field to 0 (used to check if result is available) */ for (i = 0; i < screen->mp_count; ++i) { - const unsigned b = (0x24 / 4) * i; + const unsigned b = (0x30 / 4) * i; hq->data[b + 8] = 0; } hq->sequence++; @@ -617,7 +617,7 @@ nvc0_hw_sm_query_read_data(uint32_t count[32][8], unsigned p, c; for (p = 0; p < mp_count; ++p) { - const unsigned b = (0x24 / 4) * p; + const unsigned b = (0x30 / 4) * p; for (c = 0; c < cfg->num_counters; ++c) { if (hq->data[b + 8] != hq->sequence) { @@ -815,7 +815,10 @@ nvc0_hw_sm_create_query(struct nvc0_context *nvc0, unsigned type) */ space = (4 * 4 + 4 + 4) * nvc0->screen->mp_count * sizeof(uint32_t); } else { - /* for each MP: + /* + * Note that padding is used to align memory access to 128 bits. + * + * for each MP: * [00] = MP.C0 * [04] = MP.C1 * [08] = MP.C2 @@ -825,8 +828,11 @@ nvc0_hw_sm_create_query(struct nvc0_context *nvc0, unsigned type) * [18] = MP.C6 * [1c] = MP.C7 * [20] = MP.sequence + * [24] = padding + * [28] = padding + * [2c] = padding */ - space = (8 + 1) * nvc0->screen->mp_count * sizeof(uint32_t); + space = (8 + 1 + 3) * nvc0->screen->mp_count * sizeof(uint32_t); } if (!nvc0_hw_query_allocate(nvc0, &hq->base, space)) { |