Skip to contentSkip to conversation

drm/i915: Add support for resetting the SO write pointers on gen7.

PATCH In review Eric Anholt 4 participants 6 messages Dec 30, 2011 – Jan 4, 2012

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

PATCH 1/1 drm/i915: Add support for resetting the SO write pointers on gen7.

These registers are automatically incremented by the hardware during transform feedback to track where the next streamed vertex output should go. Unlike the previous generation, which had a packet for setting the corresponding registers to a defined value, gen7 only has MI_LOAD_REGISTER_IMM to do so. That's a secure packet (since it loads an arbitrary register), so we need to do it from the kernel, and it needs to be settable atomically with the batchbuffer execution so that two clients doing transform feedback don't stomp on each others' state.

Instead of building a more complicated interface involcing setting the registers to a specific value, just set them to 0 when asked and userland can tweak its pointers accordingly.

---
 drivers/gpu/drm/i915/i915_dma.c            |    3 ++
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |   31 ++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_reg.h            |    6 +++++
 include/drm/i915_drm.h                     |    4 +++
 4 files changed, 44 insertions(+), 0 deletions(-)
drivers/gpu/drm/i915/i915_dma.c +3 -0
index a9ae374..1add685 100644
@@ -781,6 +781,9 @@ static int i915_getparam(struct drm_device *dev, void *data,
 	case I915_PARAM_HAS_RELAXED_DELTA:
 		value = 1;
 		break;
+	case I915_PARAM_HAS_GEN7_SOL_RESET:
+		value = 1;
+		break;
 	default:
 		DRM_DEBUG_DRIVER("Unknown parameter %d\n",
 				 param->param);
drivers/gpu/drm/i915/i915_gem_execbuffer.c +31 -0
index a4e4f3a..126144a 100644
@@ -971,6 +971,31 @@ i915_gem_execbuffer_retire_commands(struct drm_device *dev,
 }
 
 static int
+i915_reset_gen7_sol_offsets(struct drm_device *dev,
+			    struct intel_ring_buffer *ring)
+{
+	drm_i915_private_t *dev_priv = dev->dev_private;
+	int ret, i;
+
+	if (!IS_GEN7(dev) || ring != &dev_priv->ring[RCS])
+		return 0;
+
+	ret = intel_ring_begin(ring, 4 * 3);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < 4; i++) {
+		intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1));
+		intel_ring_emit(ring, GEN7_SO_WRITE_OFFSET(i));
+		intel_ring_emit(ring, 0);
+	}
+
+	intel_ring_advance(ring);
+
+	return 0;
+}
+
+static int
 i915_gem_do_execbuffer(struct drm_device *dev, void *data,
 		       struct drm_file *file,
 		       struct drm_i915_gem_execbuffer2 *args,
@@ -1182,6 +1207,12 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,
 		goto err;
 	}
 
+	if (args->flags & I915_EXEC_GEN7_SOL_RESET) {
+		ret = i915_reset_gen7_sol_offsets(dev, ring);
+		if (ret)
+			goto err;
+	}
+
 	trace_i915_gem_ring_dispatch(ring, seqno);
 
 	exec_start = batch_obj->gtt_offset + args->batch_start_offset;
drivers/gpu/drm/i915/i915_reg.h +6 -0
index 9d15474..54a18a4 100644
@@ -3583,4 +3583,10 @@
 #define GEN7_AUD_CNTRL_ST_A		0xE50B4
 #define GEN7_AUD_CNTRL_ST2		0xE50C0
 
+/* These are the 4 32-bit write offset registers for each stream
+ * output buffer.  It determines the offset from the
+ * 3DSTATE_SO_BUFFERs that the next streamed vertex output goes to.
+ */
+#define GEN7_SO_WRITE_OFFSET(n)		(0x5280 + (n) * 4)
+
 #endif /* _I915_REG_H_ */
include/drm/i915_drm.h +4 -0
index 28c0d11..5bb6a6a 100644
@@ -291,6 +291,7 @@ typedef struct drm_i915_irq_wait {
 #define I915_PARAM_HAS_COHERENT_RINGS	 13
 #define I915_PARAM_HAS_EXEC_CONSTANTS	 14
 #define I915_PARAM_HAS_RELAXED_DELTA	 15
+#define I915_PARAM_HAS_GEN7_SOL_RESET	 16
 
 typedef struct drm_i915_getparam {
 	int param;
@@ -653,6 +654,9 @@ struct drm_i915_gem_execbuffer2 {
 	__u64 rsvd2;
 };
 
+/** Resets the SO write offset registers for transform feedback on gen7. */
+#define I915_EXEC_GEN7_SOL_RESET	(1<<8)
+
 struct drm_i915_gem_pin {
 	/** Handle of the buffer to be pinned. */
 	__u32 handle;
Signature · 1 line
1.7.7.3
Reviewed-by on v1 1/1
#1 lore raw
Quoted 10 lines · Eric Anholt

These registers are automatically incremented by the hardware during transform feedback to track where the next streamed vertex output should go. Unlike the previous generation, which had a packet for setting the corresponding registers to a defined value, gen7 only has MI_LOAD_REGISTER_IMM to do so. That's a secure packet (since it loads an arbitrary register), so we need to do it from the kernel, and it needs to be settable atomically with the batchbuffer execution so that two clients doing transform feedback don't stomp on each others' state.

Quoted 6 lines

Instead of building a more complicated interface involcing setting the registers to a specific value, just set them to 0 when asked and userland can tweak its pointers accordingly.

---

Quoted 13 lines
 drivers/gpu/drm/i915/i915_dma.c            |    3 ++
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |   31
++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_reg.h            |    6 +++++
 include/drm/i915_drm.h                     |    4 +++
 4 files changed, 44 insertions(+), 0 deletions(-)
drivers/gpu/drm/i915/i915_dma.c +0 -0
index a9ae374..1add685 100644

drivers/gpu/drm/i915/i915_dma.c static int i915_getparam(struct drm_device *dev, void · line 787

Quoted 7 lines
drivers/gpu/drm/i915/i915_dma.c +2 -0
@@ -781,6 +781,9 @@ static int i915_getparam(struct drm_device *dev, void
 *data,
       case I915_PARAM_HAS_RELAXED_DELTA:
               value = 1;
               break;
+       case I915_PARAM_HAS_GEN7_SOL_RESET:
+               value = 1;
 

Wouldn't it be better to have:

                   value = IS_GEN7(dev);

as it is gen7+-specific item. This way, userspace could check for this support early, and avoid setting the flag on the batchbuffer in vain on pre-gen7 architectures.

Either way, it will work, so:

Signature · 2 lines
Eugeni Dodonov
<http://eugeni.dodonov.net/>

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#2 lore raw
Quoted 33 lines · Eugeni Dodonov
Quoted 20 lines · Eric Anholt

These registers are automatically incremented by the hardware during transform feedback to track where the next streamed vertex output should go. Unlike the previous generation, which had a packet for setting the corresponding registers to a defined value, gen7 only has MI_LOAD_REGISTER_IMM to do so. That's a secure packet (since it loads an arbitrary register), so we need to do it from the kernel, and it needs to be settable atomically with the batchbuffer execution so that two clients doing transform feedback don't stomp on each others' state.

Instead of building a more complicated interface involcing setting the registers to a specific value, just set them to 0 when asked and userland can tweak its pointers accordingly.

drivers/gpu/drm/i915/i915_dma.c +0 -0
index a9ae374..1add685 100644

drivers/gpu/drm/i915/i915_dma.c static int i915_getparam(struct drm_device *dev, void · line 787

Quoted 7 lines
drivers/gpu/drm/i915/i915_dma.c +2 -0
@@ -781,6 +781,9 @@ static int i915_getparam(struct drm_device *dev, void
 *data,
       case I915_PARAM_HAS_RELAXED_DELTA:
               value = 1;
               break;
+       case I915_PARAM_HAS_GEN7_SOL_RESET:
+               value = 1;
 

Wouldn't it be better to have:

                   value = IS_GEN7(dev);
Quoted 7 lines

as it is gen7+-specific item. This way, userspace could check for this support early, and avoid setting the flag on the batchbuffer in vain on pre-gen7 architectures.

Either way, it will work, so:

The flag only gets set by userland in the gen7 code, so it wouldn't change anything.

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Reviewed-by on v1 1/1
#3 lore raw
In reply to Eric Anholt
Quoted 10 lines · Eric Anholt

These registers are automatically incremented by the hardware during transform feedback to track where the next streamed vertex output should go. Unlike the previous generation, which had a packet for setting the corresponding registers to a defined value, gen7 only has MI_LOAD_REGISTER_IMM to do so. That's a secure packet (since it loads an arbitrary register), so we need to do it from the kernel, and it needs to be settable atomically with the batchbuffer execution so that two clients doing transform feedback don't stomp on each others' state.

Quoted 5 lines

Instead of building a more complicated interface involcing setting the registers to a specific value, just set them to 0 when asked and userland can tweak its pointers accordingly.

Looks great!

#4 lore raw
In reply to Eric Anholt
Quoted 113 lines · Eric Anholt

These registers are automatically incremented by the hardware during transform feedback to track where the next streamed vertex output should go. Unlike the previous generation, which had a packet for setting the corresponding registers to a defined value, gen7 only has MI_LOAD_REGISTER_IMM to do so. That's a secure packet (since it loads an arbitrary register), so we need to do it from the kernel, and it needs to be settable atomically with the batchbuffer execution so that two clients doing transform feedback don't stomp on each others' state.

Instead of building a more complicated interface involcing setting the registers to a specific value, just set them to 0 when asked and userland can tweak its pointers accordingly.

---
  drivers/gpu/drm/i915/i915_dma.c            |    3 ++
  drivers/gpu/drm/i915/i915_gem_execbuffer.c |   31 ++++++++++++++++++++++++++++
  drivers/gpu/drm/i915/i915_reg.h            |    6 +++++
  include/drm/i915_drm.h                     |    4 +++
  4 files changed, 44 insertions(+), 0 deletions(-)
drivers/gpu/drm/i915/i915_dma.c +3 -0
index a9ae374..1add685 100644
@@ -781,6 +781,9 @@ static int i915_getparam(struct drm_device *dev, void *data,
  	case I915_PARAM_HAS_RELAXED_DELTA:
  		value = 1;
  		break;
+	case I915_PARAM_HAS_GEN7_SOL_RESET:
+		value = 1;
+		break;
  	default:
  		DRM_DEBUG_DRIVER("Unknown parameter %d\n",
  				 param->param);
drivers/gpu/drm/i915/i915_gem_execbuffer.c +31 -0
index a4e4f3a..126144a 100644
@@ -971,6 +971,31 @@ i915_gem_execbuffer_retire_commands(struct drm_device *dev,
  }
 
  static int
+i915_reset_gen7_sol_offsets(struct drm_device *dev,
+			    struct intel_ring_buffer *ring)
+{
+	drm_i915_private_t *dev_priv = dev->dev_private;
+	int ret, i;
+
+	if (!IS_GEN7(dev) || ring !=&dev_priv->ring[RCS])
+		return 0;
+
+	ret = intel_ring_begin(ring, 4 * 3);
+	if (ret)
+		return ret;
+
+	for (i = 0; i<  4; i++) {
+		intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1));
+		intel_ring_emit(ring, GEN7_SO_WRITE_OFFSET(i));
+		intel_ring_emit(ring, 0);
+	}
+
+	intel_ring_advance(ring);
+
+	return 0;
+}
+
+static int
  i915_gem_do_execbuffer(struct drm_device *dev, void *data,
  		       struct drm_file *file,
  		       struct drm_i915_gem_execbuffer2 *args,
@@ -1182,6 +1207,12 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,
  		goto err;
  	}
 
+	if (args->flags&  I915_EXEC_GEN7_SOL_RESET) {
+		ret = i915_reset_gen7_sol_offsets(dev, ring);
+		if (ret)
+			goto err;
+	}
+
  	trace_i915_gem_ring_dispatch(ring, seqno);
 
  	exec_start = batch_obj->gtt_offset + args->batch_start_offset;
drivers/gpu/drm/i915/i915_reg.h +6 -0
index 9d15474..54a18a4 100644
@@ -3583,4 +3583,10 @@
  #define GEN7_AUD_CNTRL_ST_A		0xE50B4
  #define GEN7_AUD_CNTRL_ST2		0xE50C0
 
+/* These are the 4 32-bit write offset registers for each stream
+ * output buffer.  It determines the offset from the
+ * 3DSTATE_SO_BUFFERs that the next streamed vertex output goes to.
+ */
+#define GEN7_SO_WRITE_OFFSET(n)		(0x5280 + (n) * 4)
+
  #endif /* _I915_REG_H_ */
include/drm/i915_drm.h +1 -0
index 28c0d11..5bb6a6a 100644
@@ -291,6 +291,7 @@ typedef struct drm_i915_irq_wait {
  #define I915_PARAM_HAS_COHERENT_RINGS	 13
  #define I915_PARAM_HAS_EXEC_CONSTANTS	 14
  #define I915_PARAM_HAS_RELAXED_DELTA	 15
+#define I915_PARAM_HAS_GEN7_SOL_RESET	 16
 
  typedef struct drm_i915_getparam {
  	int param;

include/drm/i915_drm.h struct drm_i915_gem_execbuffer2 { · line 662

Quoted 9 lines
include/drm/i915_drm.h +3 -0
@@ -653,6 +654,9 @@ struct drm_i915_gem_execbuffer2 {
  	__u64 rsvd2;
  };
 
+/** Resets the SO write offset registers for transform feedback on gen7. */
+#define I915_EXEC_GEN7_SOL_RESET	(1<<8)
+
  struct drm_i915_gem_pin {
  	/** Handle of the buffer to be pinned. */
  	__u32 handle;

Is this something we want to carry long term wrt ABI. I really want to get context/ppgtt stuff out the door relatively soon (context should be ready to test really soon actually).

I'm totally unfamiliar with this register, but is this what you want the interface to be permanently?

Ben

#5 lore raw
Quoted 28 lines · Ben Widawsky
Quoted 16 lines · Eric Anholt

These registers are automatically incremented by the hardware during transform feedback to track where the next streamed vertex output should go. Unlike the previous generation, which had a packet for setting the corresponding registers to a defined value, gen7 only has MI_LOAD_REGISTER_IMM to do so. That's a secure packet (since it loads an arbitrary register), so we need to do it from the kernel, and it needs to be settable atomically with the batchbuffer execution so that two clients doing transform feedback don't stomp on each others' state.

Instead of building a more complicated interface involcing setting the registers to a specific value, just set them to 0 when asked and userland can tweak its pointers accordingly.

include/drm/i915_drm.h struct drm_i915_gem_execbuffer2 { · line 662

Quoted 6 lines
include/drm/i915_drm.h +3 -0
+/** Resets the SO write offset registers for transform feedback on gen7. */
+#define I915_EXEC_GEN7_SOL_RESET	(1<<8)
+
  struct drm_i915_gem_pin {
  	/** Handle of the buffer to be pinned. */
  	__u32 handle;

Is this something we want to carry long term wrt ABI. I really want to get context/ppgtt stuff out the door relatively soon (context should be ready to test really soon actually).

Quoted 2 lines

I'm totally unfamiliar with this register, but is this what you want the interface to be permanently?

Ultimately, we want this ABI plus contexts. Until then, we can live with this ABI alone.

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx