Skip to contentSkip to conversation

drm/amdgpu: simplify padding calculations

PATCH In review Tuikov, Luben 2 participants 8 messages Oct 24, 2019 – Oct 28, 2019 5 messages not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/amdgpu: simplify padding calculations

Simplify padding calculations.

---
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c  |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 14 +++++++++-----
 5 files changed, 17 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +2 -2
index c45304f1047c..1ea9e18d6f08 100644
@@ -228,7 +228,7 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
 	u32 extra_bits = vmid & 0xf;
 
 	/* IB packet must end on a 8 DW boundary */
-	cik_sdma_ring_insert_nop(ring, (12 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	cik_sdma_ring_insert_nop(ring, (4-lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
 	amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
@@ -811,7 +811,7 @@ static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +2 -2
index a10175838013..d340f179401a 100644
@@ -255,7 +255,7 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
 	/* IB packet must end on a 8 DW boundary */
-	sdma_v2_4_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v2_4_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -750,7 +750,7 @@ static void sdma_v2_4_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +2 -2
index 5f4e2c616241..5c3c310188b6 100644
@@ -429,7 +429,7 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
 	/* IB packet must end on a 8 DW boundary */
-	sdma_v3_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v3_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1021,7 +1021,7 @@ static void sdma_v3_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -2
index 45bd538ba97e..7c71c88e38a4 100644
@@ -698,7 +698,7 @@ static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
 	/* IB packet must end on a 8 DW boundary */
-	sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v4_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1580,7 +1580,7 @@ static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +9 -5
index 0c41b4fdc58b..67ede9e4df01 100644
@@ -382,8 +382,12 @@ static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 	uint64_t csa_mc_addr = amdgpu_sdma_get_csa_mc_addr(ring, vmid);
 
-	/* IB packet must end on a 8 DW boundary */
-	sdma_v5_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	/* An IB packet must end on a 8 DW boundary--the next dword must
+	 * be on a 8-dword boundary. Our IB below is 6 dwords long,
+	 * thus add x number of NOPs, such that
+	 * wptr + 6 + x = 8k, k >= 0.
+	 */
+	sdma_v5_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1086,10 +1090,10 @@ static void sdma_v5_0_vm_set_pte_pde(struct amdgpu_ib *ib,
 }
 
 /**
- * sdma_v5_0_ring_pad_ib - pad the IB to the required number of dw
- *
+ * sdma_v5_0_ring_pad_ib - pad the IB
  * @ib: indirect buffer to fill with padding
  *
+ * Pad the IB with NOPs to a boundary multiple of 8.
  */
 static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
 {
@@ -1097,7 +1101,7 @@ static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 0x7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
Signature · 6 lines
2.23.0.385.gbc12974a89

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): [email protected]

#2 lore raw

Am 25.10.19 um 01:44 schrieb Tuikov, Luben:

Quoted 104 lines

Simplify padding calculations.

---
  drivers/gpu/drm/amd/amdgpu/cik_sdma.c  |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 14 +++++++++-----
  5 files changed, 17 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +2 -2
index c45304f1047c..1ea9e18d6f08 100644
@@ -228,7 +228,7 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
  	u32 extra_bits = vmid & 0xf;
 
  	/* IB packet must end on a 8 DW boundary */
-	cik_sdma_ring_insert_nop(ring, (12 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	cik_sdma_ring_insert_nop(ring, (4-lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
  	amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
@@ -811,7 +811,7 @@ static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +2 -2
index a10175838013..d340f179401a 100644
@@ -255,7 +255,7 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
  	/* IB packet must end on a 8 DW boundary */
-	sdma_v2_4_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v2_4_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -750,7 +750,7 @@ static void sdma_v2_4_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +2 -2
index 5f4e2c616241..5c3c310188b6 100644
@@ -429,7 +429,7 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
  	/* IB packet must end on a 8 DW boundary */
-	sdma_v3_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v3_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1021,7 +1021,7 @@ static void sdma_v3_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -2
index 45bd538ba97e..7c71c88e38a4 100644
@@ -698,7 +698,7 @@ static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
  	/* IB packet must end on a 8 DW boundary */
-	sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v4_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1580,7 +1580,7 @@ static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +0 -0
index 0c41b4fdc58b..67ede9e4df01 100644

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 386

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -2
@@ -382,8 +382,12 @@ static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
  	uint64_t csa_mc_addr = amdgpu_sdma_get_csa_mc_addr(ring, vmid);
 
-	/* IB packet must end on a 8 DW boundary */
-	sdma_v5_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	/* An IB packet must end on a 8 DW boundary--the next dword must
+	 * be on a 8-dword boundary. Our IB below is 6 dwords long,

That should probably read "our IB command" or "our IB packet".

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 388

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -0
+	 * thus add x number of NOPs, such that
+	 * wptr + 6 + x = 8k, k >= 0.

Maybe write (wrptr + 6 + x) % 8 = 0 to make that more obvious.

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 390

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -0
+	 */
+	sdma_v5_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);

Spaces between operators please.

Apart form that looks good to me,
Christian.

Quoted 17 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -2
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1086,10 +1090,10 @@ static void sdma_v5_0_vm_set_pte_pde(struct amdgpu_ib *ib,
  }
 
  /**
- * sdma_v5_0_ring_pad_ib - pad the IB to the required number of dw
- *
+ * sdma_v5_0_ring_pad_ib - pad the IB
   * @ib: indirect buffer to fill with padding
   *
+ * Pad the IB with NOPs to a boundary multiple of 8.
   */
  static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
  {

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib · line 1107

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +1 -1
@@ -1097,7 +1101,7 @@ static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 0x7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 45e07677-717a-2c01-369e-1dbd19ce8e74-5C7GfCeVMHo@public.gmane.org

#4 lore raw
Quoted 108 lines · Christian

Am 25.10.19 um 01:44 schrieb Tuikov, Luben:

Quoted 107 lines

Simplify padding calculations.

---
  drivers/gpu/drm/amd/amdgpu/cik_sdma.c  |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 14 +++++++++-----
  5 files changed, 17 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +2 -2
index c45304f1047c..1ea9e18d6f08 100644
@@ -228,7 +228,7 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
  	u32 extra_bits = vmid & 0xf;
 
  	/* IB packet must end on a 8 DW boundary */
-	cik_sdma_ring_insert_nop(ring, (12 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	cik_sdma_ring_insert_nop(ring, (4-lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
  	amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
@@ -811,7 +811,7 @@ static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +2 -2
index a10175838013..d340f179401a 100644
@@ -255,7 +255,7 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
  	/* IB packet must end on a 8 DW boundary */
-	sdma_v2_4_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v2_4_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -750,7 +750,7 @@ static void sdma_v2_4_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +2 -2
index 5f4e2c616241..5c3c310188b6 100644
@@ -429,7 +429,7 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
  	/* IB packet must end on a 8 DW boundary */
-	sdma_v3_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v3_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1021,7 +1021,7 @@ static void sdma_v3_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -2
index 45bd538ba97e..7c71c88e38a4 100644
@@ -698,7 +698,7 @@ static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
  	/* IB packet must end on a 8 DW boundary */
-	sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v4_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1580,7 +1580,7 @@ static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +0 -0
index 0c41b4fdc58b..67ede9e4df01 100644
@@ -382,8 +382,12 @@ static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
  	uint64_t csa_mc_addr = amdgpu_sdma_get_csa_mc_addr(ring, vmid);
 
Quoted 6 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 386

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -2
-	/* IB packet must end on a 8 DW boundary */
-	sdma_v5_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	/* An IB packet must end on a 8 DW boundary--the next dword must
+	 * be on a 8-dword boundary. Our IB below is 6 dwords long,

That should probably read "our IB command" or "our IB packet".

Done.

Quoted 5 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 388

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -0
+	 * thus add x number of NOPs, such that
+	 * wptr + 6 + x = 8k, k >= 0.

Maybe write (wrptr + 6 + x) % 8 = 0 to make that more obvious.

That's less obvious. In my mind I always translate it to a congruence expression. It's what I used to derive the algebraic simplifications of this patch.

I'll add the highly cryptic "(wrptr + 6 + x) % 8 = 0" to make you happy, below the algebraic expression already there.

Quoted 5 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 390

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -0
+	 */
+	sdma_v5_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);

Spaces between operators please.

Yeah, I wanted to put an emphasis on the AND-op, but I'll add them.

Regards,
Luben

Quoted 24 lines

Apart form that looks good to me,
Christian.

Quoted 20 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -2
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1086,10 +1090,10 @@ static void sdma_v5_0_vm_set_pte_pde(struct amdgpu_ib *ib,
  }
 
  /**
- * sdma_v5_0_ring_pad_ib - pad the IB to the required number of dw
- *
+ * sdma_v5_0_ring_pad_ib - pad the IB
   * @ib: indirect buffer to fill with padding
   *
+ * Pad the IB with NOPs to a boundary multiple of 8.
   */
  static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
  {
@@ -1097,7 +1101,7 @@ static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
Quoted 6 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib · line 1107

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +1 -1
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 0x7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): ed713bfe-2032-b56a-3e29-f61f48aeeb3b-5C7GfCeVMHo@public.gmane.org

#6 lore raw

PATCH 1/1 drm/amdgpu: simplify padding calculations

Simplify padding calculations.

---
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c  |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 17 ++++++++++++-----
 5 files changed, 20 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +2 -2
index c45304f1047c..1ea9e18d6f08 100644
@@ -228,7 +228,7 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
 	u32 extra_bits = vmid & 0xf;
 
 	/* IB packet must end on a 8 DW boundary */
-	cik_sdma_ring_insert_nop(ring, (12 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	cik_sdma_ring_insert_nop(ring, (4-lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
 	amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
@@ -811,7 +811,7 @@ static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +2 -2
index a10175838013..d340f179401a 100644
@@ -255,7 +255,7 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
 	/* IB packet must end on a 8 DW boundary */
-	sdma_v2_4_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v2_4_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -750,7 +750,7 @@ static void sdma_v2_4_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +2 -2
index 5f4e2c616241..5c3c310188b6 100644
@@ -429,7 +429,7 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
 	/* IB packet must end on a 8 DW boundary */
-	sdma_v3_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v3_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1021,7 +1021,7 @@ static void sdma_v3_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -2
index 45bd538ba97e..7c71c88e38a4 100644
@@ -698,7 +698,7 @@ static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
 	/* IB packet must end on a 8 DW boundary */
-	sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v4_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1580,7 +1580,7 @@ static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +12 -5
index 0c41b4fdc58b..d117bde3f29a 100644
@@ -382,8 +382,15 @@ static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 	uint64_t csa_mc_addr = amdgpu_sdma_get_csa_mc_addr(ring, vmid);
 
-	/* IB packet must end on a 8 DW boundary */
-	sdma_v5_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	/* An IB packet must end on a 8 DW boundary--the next dword
+	 * must be on a 8-dword boundary. Our IB packet below is 6
+	 * dwords long, thus add x number of NOPs, such that, in
+	 * modular arithmetic,
+	 * wptr + 6 + x = 8k, k >= 0, which in C is,
+	 * (wptr + 6 + x) % 8 = 0.
+	 * The expression below, is a solution of x.
+	 */
+	sdma_v5_0_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1086,10 +1093,10 @@ static void sdma_v5_0_vm_set_pte_pde(struct amdgpu_ib *ib,
 }
 
 /**
- * sdma_v5_0_ring_pad_ib - pad the IB to the required number of dw
- *
+ * sdma_v5_0_ring_pad_ib - pad the IB
  * @ib: indirect buffer to fill with padding
  *
+ * Pad the IB with NOPs to a boundary multiple of 8.
  */
 static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
 {
@@ -1097,7 +1104,7 @@ static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 0x7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
Signature · 6 lines
2.23.0.385.gbc12974a89

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#7 lore raw
In reply to a message not in the archive

Am 25.10.19 um 23:51 schrieb Tuikov, Luben:

Quoted 119 lines · Christian
Quoted 108 lines

Am 25.10.19 um 01:44 schrieb Tuikov, Luben:

Quoted 107 lines

Simplify padding calculations.

---
   drivers/gpu/drm/amd/amdgpu/cik_sdma.c  |  4 ++--
   drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c |  4 ++--
   drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c |  4 ++--
   drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c |  4 ++--
   drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 14 +++++++++-----
   5 files changed, 17 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +2 -2
index c45304f1047c..1ea9e18d6f08 100644
@@ -228,7 +228,7 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
   	u32 extra_bits = vmid & 0xf;
 
   	/* IB packet must end on a 8 DW boundary */
-	cik_sdma_ring_insert_nop(ring, (12 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	cik_sdma_ring_insert_nop(ring, (4-lower_32_bits(ring->wptr)) & 7);
 
   	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
   	amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
@@ -811,7 +811,7 @@ static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
   	u32 pad_count;
   	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
   	for (i = 0; i < pad_count; i++)
   		if (sdma && sdma->burst_nop && (i == 0))
   			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +2 -2
index a10175838013..d340f179401a 100644
@@ -255,7 +255,7 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
   	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
   	/* IB packet must end on a 8 DW boundary */
-	sdma_v2_4_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v2_4_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
   	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
   			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -750,7 +750,7 @@ static void sdma_v2_4_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
   	u32 pad_count;
   	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
   	for (i = 0; i < pad_count; i++)
   		if (sdma && sdma->burst_nop && (i == 0))
   			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +2 -2
index 5f4e2c616241..5c3c310188b6 100644
@@ -429,7 +429,7 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
   	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
   	/* IB packet must end on a 8 DW boundary */
-	sdma_v3_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v3_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
   	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
   			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1021,7 +1021,7 @@ static void sdma_v3_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
   	u32 pad_count;
   	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
   	for (i = 0; i < pad_count; i++)
   		if (sdma && sdma->burst_nop && (i == 0))
   			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -2
index 45bd538ba97e..7c71c88e38a4 100644
@@ -698,7 +698,7 @@ static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
   	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
   	/* IB packet must end on a 8 DW boundary */
-	sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v4_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
   	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
   			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1580,7 +1580,7 @@ static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
   	u32 pad_count;
   	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
   	for (i = 0; i < pad_count; i++)
   		if (sdma && sdma->burst_nop && (i == 0))
   			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +0 -0
index 0c41b4fdc58b..67ede9e4df01 100644
@@ -382,8 +382,12 @@ static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
   	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
   	uint64_t csa_mc_addr = amdgpu_sdma_get_csa_mc_addr(ring, vmid);
 
Quoted 5 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 386

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -2
-	/* IB packet must end on a 8 DW boundary */
-	sdma_v5_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	/* An IB packet must end on a 8 DW boundary--the next dword must
+	 * be on a 8-dword boundary. Our IB below is 6 dwords long,

That should probably read "our IB command" or "our IB packet".

Done.

Quoted 3 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 388

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -0
+	 * thus add x number of NOPs, such that
+	 * wptr + 6 + x = 8k, k >= 0.

Maybe write (wrptr + 6 + x) % 8 = 0 to make that more obvious.

Quoted 6 lines

That's less obvious. In my mind I always translate it to a congruence expression. It's what I used to derive the algebraic simplifications of this patch.

I'll add the highly cryptic "(wrptr + 6 + x) % 8 = 0" to make you happy, below the algebraic expression already there.

The problem I see here is that people who are used to GPUs will think that k is a float and not immediately get what's meant here.

Quoted 4 lines
Quoted 3 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 390

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -0
+	 */
+	sdma_v5_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);

Spaces between operators please.

Yeah, I wanted to put an emphasis on the AND-op, but I'll add them.

What would be really nice to have is to keep the 10 as number of DW in the IB packet around.

Maybe add an "static inline amdgpu_ring_unalign(ring, 10, 8)" helper or something like that, the insert_nop is a ring callback anyway IIRC.

Regards,
Christian.

Quoted 28 lines

Regards,
Luben

Quoted 23 lines

Apart form that looks good to me,
Christian.

Quoted 20 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -2
   	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
   			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1086,10 +1090,10 @@ static void sdma_v5_0_vm_set_pte_pde(struct amdgpu_ib *ib,
   }
 
   /**
- * sdma_v5_0_ring_pad_ib - pad the IB to the required number of dw
- *
+ * sdma_v5_0_ring_pad_ib - pad the IB
    * @ib: indirect buffer to fill with padding
    *
+ * Pad the IB with NOPs to a boundary multiple of 8.
    */
   static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
   {
@@ -1097,7 +1101,7 @@ static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
   	u32 pad_count;
   	int i;
 
Quoted 5 lines
Quoted 5 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib · line 1107

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +1 -1
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 0x7;
   	for (i = 0; i < pad_count; i++)
   		if (sdma && sdma->burst_nop && (i == 0))
   			ib->ptr[ib->length_dw++] =

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): db4d6ece-2cec-a465-0639-d75142621dda-5C7GfCeVMHo@public.gmane.org

#9 lore raw
Quoted 130 lines · Christian

Am 25.10.19 um 23:51 schrieb Tuikov, Luben:

Quoted 119 lines · Christian
Quoted 108 lines

Am 25.10.19 um 01:44 schrieb Tuikov, Luben:

Quoted 107 lines

Simplify padding calculations.

---
   drivers/gpu/drm/amd/amdgpu/cik_sdma.c  |  4 ++--
   drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c |  4 ++--
   drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c |  4 ++--
   drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c |  4 ++--
   drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 14 +++++++++-----
   5 files changed, 17 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +2 -2
index c45304f1047c..1ea9e18d6f08 100644
@@ -228,7 +228,7 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
   	u32 extra_bits = vmid & 0xf;
 
   	/* IB packet must end on a 8 DW boundary */
-	cik_sdma_ring_insert_nop(ring, (12 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	cik_sdma_ring_insert_nop(ring, (4-lower_32_bits(ring->wptr)) & 7);
 
   	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
   	amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
@@ -811,7 +811,7 @@ static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
   	u32 pad_count;
   	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
   	for (i = 0; i < pad_count; i++)
   		if (sdma && sdma->burst_nop && (i == 0))
   			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +2 -2
index a10175838013..d340f179401a 100644
@@ -255,7 +255,7 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
   	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
   	/* IB packet must end on a 8 DW boundary */
-	sdma_v2_4_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v2_4_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
   	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
   			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -750,7 +750,7 @@ static void sdma_v2_4_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
   	u32 pad_count;
   	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
   	for (i = 0; i < pad_count; i++)
   		if (sdma && sdma->burst_nop && (i == 0))
   			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +2 -2
index 5f4e2c616241..5c3c310188b6 100644
@@ -429,7 +429,7 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
   	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
   	/* IB packet must end on a 8 DW boundary */
-	sdma_v3_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v3_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
   	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
   			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1021,7 +1021,7 @@ static void sdma_v3_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
   	u32 pad_count;
   	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
   	for (i = 0; i < pad_count; i++)
   		if (sdma && sdma->burst_nop && (i == 0))
   			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -2
index 45bd538ba97e..7c71c88e38a4 100644
@@ -698,7 +698,7 @@ static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
   	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
   	/* IB packet must end on a 8 DW boundary */
-	sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v4_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);
 
   	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
   			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1580,7 +1580,7 @@ static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
   	u32 pad_count;
   	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
   	for (i = 0; i < pad_count; i++)
   		if (sdma && sdma->burst_nop && (i == 0))
   			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +0 -0
index 0c41b4fdc58b..67ede9e4df01 100644
@@ -382,8 +382,12 @@ static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
   	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
   	uint64_t csa_mc_addr = amdgpu_sdma_get_csa_mc_addr(ring, vmid);
 
Quoted 5 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 386

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -2
-	/* IB packet must end on a 8 DW boundary */
-	sdma_v5_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	/* An IB packet must end on a 8 DW boundary--the next dword must
+	 * be on a 8-dword boundary. Our IB below is 6 dwords long,

That should probably read "our IB command" or "our IB packet".

Done.

Quoted 3 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 388

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -0
+	 * thus add x number of NOPs, such that
+	 * wptr + 6 + x = 8k, k >= 0.

Maybe write (wrptr + 6 + x) % 8 = 0 to make that more obvious.

Quoted 6 lines

That's less obvious. In my mind I always translate it to a congruence expression. It's what I used to derive the algebraic simplifications of this patch.

I'll add the highly cryptic "(wrptr + 6 + x) % 8 = 0" to make you happy, below the algebraic expression already there.

The problem I see here is that people who are used to GPUs will think that k is a float and not immediately get what's meant here.

Quoted 7 lines
Quoted 4 lines
Quoted 3 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring, · line 390

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -0
+	 */
+	sdma_v5_0_ring_insert_nop(ring, (2-lower_32_bits(ring->wptr)) & 7);

Spaces between operators please.

Yeah, I wanted to put an emphasis on the AND-op, but I'll add them.

What would be really nice to have is to keep the 10 as number of DW in the IB packet around.

Why? Why 10?

As far as I can see, THE IB packet is 6 dwords, we want to pad with some number of dwords, in order to make sure that the wptr + 6, ends up on a boundary multiple of 8. I.e. we want to solve exactly this, for x:

wptr + 6 + x = 8k, k >= 0.

x = -(wtpr + 6) = -6 - wptr = 2 - wptr (mod 8)

Also note that, 10 = 2 = -6 (mod 8).

I believe the patch is good as is.

Did you see v2 of the patch?

Regards,
Luben

Quoted 35 lines

Maybe add an "static inline amdgpu_ring_unalign(ring, 10, 8)" helper or something like that, the insert_nop is a ring callback anyway IIRC.

Regards,
Christian.

Quoted 28 lines

Regards,
Luben

Quoted 23 lines

Apart form that looks good to me,
Christian.

Quoted 20 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +2 -2
   	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
   			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1086,10 +1090,10 @@ static void sdma_v5_0_vm_set_pte_pde(struct amdgpu_ib *ib,
   }
 
   /**
- * sdma_v5_0_ring_pad_ib - pad the IB to the required number of dw
- *
+ * sdma_v5_0_ring_pad_ib - pad the IB
    * @ib: indirect buffer to fill with padding
    *
+ * Pad the IB with NOPs to a boundary multiple of 8.
    */
   static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
   {
@@ -1097,7 +1101,7 @@ static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
   	u32 pad_count;
   	int i;
 
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib · line 1107

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +1 -1
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 0x7;
   	for (i = 0; i < pad_count; i++)
   		if (sdma && sdma->burst_nop && (i == 0))
   			ib->ptr[ib->length_dw++] =

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#10 lore raw

PATCH 1/1 drm/amdgpu: simplify padding calculations (v2)

In reply to a message not in the archive

Simplify padding calculations.

v2: Comment update and spacing.

---
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c  |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 17 ++++++++++++-----
 5 files changed, 20 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +2 -2
index c45304f1047c..4af9acc2dc4f 100644
@@ -228,7 +228,7 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
 	u32 extra_bits = vmid & 0xf;
 
 	/* IB packet must end on a 8 DW boundary */
-	cik_sdma_ring_insert_nop(ring, (12 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	cik_sdma_ring_insert_nop(ring, (4 - lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
 	amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
@@ -811,7 +811,7 @@ static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +2 -2
index a10175838013..b6af67f6f214 100644
@@ -255,7 +255,7 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
 	/* IB packet must end on a 8 DW boundary */
-	sdma_v2_4_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v2_4_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -750,7 +750,7 @@ static void sdma_v2_4_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +2 -2
index 5f4e2c616241..cd3ebed46d05 100644
@@ -429,7 +429,7 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
 	/* IB packet must end on a 8 DW boundary */
-	sdma_v3_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v3_0_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1021,7 +1021,7 @@ static void sdma_v3_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -2
index 45bd538ba97e..8ce15056ee4f 100644
@@ -698,7 +698,7 @@ static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
 	/* IB packet must end on a 8 DW boundary */
-	sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v4_0_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1580,7 +1580,7 @@ static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +12 -5
index 0c41b4fdc58b..d117bde3f29a 100644
@@ -382,8 +382,15 @@ static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 	uint64_t csa_mc_addr = amdgpu_sdma_get_csa_mc_addr(ring, vmid);
 
-	/* IB packet must end on a 8 DW boundary */
-	sdma_v5_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	/* An IB packet must end on a 8 DW boundary--the next dword
+	 * must be on a 8-dword boundary. Our IB packet below is 6
+	 * dwords long, thus add x number of NOPs, such that, in
+	 * modular arithmetic,
+	 * wptr + 6 + x = 8k, k >= 0, which in C is,
+	 * (wptr + 6 + x) % 8 = 0.
+	 * The expression below, is a solution of x.
+	 */
+	sdma_v5_0_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
 
 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1086,10 +1093,10 @@ static void sdma_v5_0_vm_set_pte_pde(struct amdgpu_ib *ib,
 }
 
 /**
- * sdma_v5_0_ring_pad_ib - pad the IB to the required number of dw
- *
+ * sdma_v5_0_ring_pad_ib - pad the IB
  * @ib: indirect buffer to fill with padding
  *
+ * Pad the IB with NOPs to a boundary multiple of 8.
  */
 static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
 {
@@ -1097,7 +1104,7 @@ static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
 	u32 pad_count;
 	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 0x7;
 	for (i = 0; i < pad_count; i++)
 		if (sdma && sdma->burst_nop && (i == 0))
 			ib->ptr[ib->length_dw++] =
Signature · 6 lines
2.23.0.385.gbc12974a89

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): [email protected]

Reviewed-by Christian König
#12 lore raw

Re: [PATCH] drm/amdgpu: simplify padding calculations (v2)

Am 26.10.19 um 00:41 schrieb Tuikov, Luben:

Quoted 5 lines

Simplify padding calculations.

v2: Comment update and spacing.

Quoted 132 lines
---
  drivers/gpu/drm/amd/amdgpu/cik_sdma.c  |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c |  4 ++--
  drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 17 ++++++++++++-----
  5 files changed, 20 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +2 -2
index c45304f1047c..4af9acc2dc4f 100644
@@ -228,7 +228,7 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
  	u32 extra_bits = vmid & 0xf;
 
  	/* IB packet must end on a 8 DW boundary */
-	cik_sdma_ring_insert_nop(ring, (12 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	cik_sdma_ring_insert_nop(ring, (4 - lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
  	amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
@@ -811,7 +811,7 @@ static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +2 -2
index a10175838013..b6af67f6f214 100644
@@ -255,7 +255,7 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
  	/* IB packet must end on a 8 DW boundary */
-	sdma_v2_4_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v2_4_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -750,7 +750,7 @@ static void sdma_v2_4_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +2 -2
index 5f4e2c616241..cd3ebed46d05 100644
@@ -429,7 +429,7 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
  	/* IB packet must end on a 8 DW boundary */
-	sdma_v3_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v3_0_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1021,7 +1021,7 @@ static void sdma_v3_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -2
index 45bd538ba97e..8ce15056ee4f 100644
@@ -698,7 +698,7 @@ static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
 
  	/* IB packet must end on a 8 DW boundary */
-	sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	sdma_v4_0_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1580,7 +1580,7 @@ static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +11 -4
index 0c41b4fdc58b..d117bde3f29a 100644
@@ -382,8 +382,15 @@ static void sdma_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
  	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
  	uint64_t csa_mc_addr = amdgpu_sdma_get_csa_mc_addr(ring, vmid);
 
-	/* IB packet must end on a 8 DW boundary */
-	sdma_v5_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
+	/* An IB packet must end on a 8 DW boundary--the next dword
+	 * must be on a 8-dword boundary. Our IB packet below is 6
+	 * dwords long, thus add x number of NOPs, such that, in
+	 * modular arithmetic,
+	 * wptr + 6 + x = 8k, k >= 0, which in C is,
+	 * (wptr + 6 + x) % 8 = 0.
+	 * The expression below, is a solution of x.
+	 */
+	sdma_v5_0_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
 
  	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
  			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
@@ -1086,10 +1093,10 @@ static void sdma_v5_0_vm_set_pte_pde(struct amdgpu_ib *ib,
  }
 
  /**
- * sdma_v5_0_ring_pad_ib - pad the IB to the required number of dw
- *
+ * sdma_v5_0_ring_pad_ib - pad the IB
   * @ib: indirect buffer to fill with padding
   *
+ * Pad the IB with NOPs to a boundary multiple of 8.
   */
  static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
  {

drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib · line 1110

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +1 -1
@@ -1097,7 +1104,7 @@ static void sdma_v5_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib
  	u32 pad_count;
  	int i;
 
-	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
+	pad_count = (-ib->length_dw) & 0x7;
  	for (i = 0; i < pad_count; i++)
  		if (sdma && sdma->burst_nop && (i == 0))
  			ib->ptr[ib->length_dw++] =

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx