Avoid constant register reloads while emitting IBs by using a local write
pointer and only updating the size at the end of each helper.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursu...@igalia.com>
---
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 106 +++++++++++++++----------
1 file changed, 64 insertions(+), 42 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
b/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
index 7dc67a22a7a0..ab012592a276 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
@@ -1155,16 +1155,19 @@ static void sdma_v5_0_vm_copy_pte(struct amdgpu_ib *ib,
uint64_t pe, uint64_t src,
unsigned count)
{
+ u32 *ptr = &ib->ptr[ib->length_dw];
unsigned bytes = count * 8;
- ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
- SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
- ib->ptr[ib->length_dw++] = bytes - 1;
- ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
- ib->ptr[ib->length_dw++] = lower_32_bits(src);
- ib->ptr[ib->length_dw++] = upper_32_bits(src);
- ib->ptr[ib->length_dw++] = lower_32_bits(pe);
- ib->ptr[ib->length_dw++] = upper_32_bits(pe);
+ *ptr++ = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
+ SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
+ *ptr++ = bytes - 1;
+ *ptr++ = 0; /* src/dst endian swap */
+ *ptr++ = lower_32_bits(src);
+ *ptr++ = upper_32_bits(src);
+ *ptr++ = lower_32_bits(pe);
+ *ptr++ = upper_32_bits(pe);
+
+ ib->length_dw = ptr - ib->ptr;
}
@@ -1183,18 +1186,21 @@ static void sdma_v5_0_vm_write_pte(struct amdgpu_ib
*ib, uint64_t pe,
uint64_t value, unsigned count,
uint32_t incr)
{
+ u32 *ptr = &ib->ptr[ib->length_dw];
unsigned ndw = count * 2;
- ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
- SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
- ib->ptr[ib->length_dw++] = lower_32_bits(pe);
- ib->ptr[ib->length_dw++] = upper_32_bits(pe);
- ib->ptr[ib->length_dw++] = ndw - 1;
+ *ptr++ = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
+ SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
+ *ptr++ = lower_32_bits(pe);
+ *ptr++ = upper_32_bits(pe);
+ *ptr++ = ndw - 1;
for (; ndw > 0; ndw -= 2) {
- ib->ptr[ib->length_dw++] = lower_32_bits(value);
- ib->ptr[ib->length_dw++] = upper_32_bits(value);
+ *ptr++ = lower_32_bits(value);
+ *ptr++ = upper_32_bits(value);
value += incr;
}
+
+ ib->length_dw = ptr - ib->ptr;
}
/**
@@ -1214,17 +1220,21 @@ static void sdma_v5_0_vm_set_pte_pde(struct amdgpu_ib
*ib,
uint64_t addr, unsigned count,
uint32_t incr, uint64_t flags)
{
+ u32 *ptr = &ib->ptr[ib->length_dw];
+
/* for physically contiguous pages (vram) */
- ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_PTEPDE);
- ib->ptr[ib->length_dw++] = lower_32_bits(pe); /* dst addr */
- ib->ptr[ib->length_dw++] = upper_32_bits(pe);
- ib->ptr[ib->length_dw++] = lower_32_bits(flags); /* mask */
- ib->ptr[ib->length_dw++] = upper_32_bits(flags);
- ib->ptr[ib->length_dw++] = lower_32_bits(addr); /* value */
- ib->ptr[ib->length_dw++] = upper_32_bits(addr);
- ib->ptr[ib->length_dw++] = incr; /* increment size */
- ib->ptr[ib->length_dw++] = 0;
- ib->ptr[ib->length_dw++] = count - 1; /* number of entries */
+ *ptr++ = SDMA_PKT_HEADER_OP(SDMA_OP_PTEPDE);
+ *ptr++ = lower_32_bits(pe); /* dst addr */
+ *ptr++ = upper_32_bits(pe);
+ *ptr++ = lower_32_bits(flags); /* mask */
+ *ptr++ = upper_32_bits(flags);
+ *ptr++ = lower_32_bits(addr); /* value */
+ *ptr++ = upper_32_bits(addr);
+ *ptr++ = incr; /* increment size */
+ *ptr++ = 0;
+ *ptr++ = count - 1; /* number of entries */
+
+ ib->length_dw = ptr - ib->ptr;
}
/**
@@ -1237,18 +1247,22 @@ static void sdma_v5_0_vm_set_pte_pde(struct amdgpu_ib
*ib,
static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib
*ib)
{
struct amdgpu_sdma_instance *sdma =
amdgpu_sdma_get_instance_from_ring(ring);
+ u32 *ptr = &ib->ptr[ib->length_dw];
u32 pad_count;
int i;
pad_count = (-ib->length_dw) & 0x7;
+ if (!pad_count)
+ return;
+
for (i = 0; i < pad_count; i++)
if (sdma && sdma->burst_nop && (i == 0))
- ib->ptr[ib->length_dw++] =
- SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
- SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
+ *ptr++ = SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
+ SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
else
- ib->ptr[ib->length_dw++] =
- SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
+ *ptr++ = SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
+
+ ib->length_dw += pad_count;
}
@@ -2021,15 +2035,19 @@ static void sdma_v5_0_emit_copy_buffer(struct amdgpu_ib
*ib,
uint32_t byte_count,
uint32_t copy_flags)
{
- ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
+ u32 *ptr = &ib->ptr[ib->length_dw];
+
+ *ptr++ = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR) |
SDMA_PKT_COPY_LINEAR_HEADER_TMZ((copy_flags &
AMDGPU_COPY_FLAGS_TMZ) ? 1 : 0);
- ib->ptr[ib->length_dw++] = byte_count - 1;
- ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
- ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
- ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
- ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
- ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
+ *ptr++ = byte_count - 1;
+ *ptr++ = 0; /* src/dst endian swap */
+ *ptr++ = lower_32_bits(src_offset);
+ *ptr++ = upper_32_bits(src_offset);
+ *ptr++ = lower_32_bits(dst_offset);
+ *ptr++ = upper_32_bits(dst_offset);
+
+ ib->length_dw = ptr - ib->ptr;
}
/**
@@ -2047,11 +2065,15 @@ static void sdma_v5_0_emit_fill_buffer(struct amdgpu_ib
*ib,
uint64_t dst_offset,
uint32_t byte_count)
{
- ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
- ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
- ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
- ib->ptr[ib->length_dw++] = src_data;
- ib->ptr[ib->length_dw++] = byte_count - 1;
+ u32 *ptr = &ib->ptr[ib->length_dw];
+
+ *ptr++ = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
+ *ptr++ = lower_32_bits(dst_offset);
+ *ptr++ = upper_32_bits(dst_offset);
+ *ptr++ = src_data;
+ *ptr++ = byte_count - 1;
+
+ ib->length_dw = ptr - ib->ptr;
}
static const struct amdgpu_buffer_funcs sdma_v5_0_buffer_funcs = {
--
2.48.0
