freedreno/a6xx: Implement PIPE_QUERY_PRIMITIVES_GENERATED for GS

When we don't have streamout enabled, we have to read this register to
get the number of primitives emitted.

Signed-off-by: Kristian H. Kristensen <[email protected]>

3 files changed, 68 insertions, 0 deletions

diff --git a/src/freedreno/registers/a6xx.xml b/src/freedreno/registers/a6xx.xml
index 4d83a5dd2cf..b7cfecdc121 100644
--- a/src/freedreno/registers/a6xx.xml
+++ b/src/freedreno/registers/a6xx.xml
@@ -1372,6 +1372,38 @@ to upconvert to 32b float internally?
 	<reg32 offset="0x050A" name="RBBM_PERFCTR_RBBM_SEL_3"/>
 	<reg32 offset="0x050B" name="RBBM_PERFCTR_GPU_BUSY_MASKED"/>
 	<reg32 offset="0x0533" name="RBBM_ISDB_CNT"/>
+
+	<!---
+	    This block of registers aren't tied to perf counters. They
+	    count various geometry stats, for example number of
+	    vertices in, number of primnitives assembled etc.
+	-->
+
+	<reg32 offset="0x0540" name="RBBM_PRIMCTR_0_LO"/>
+	<reg32 offset="0x0541" name="RBBM_PRIMCTR_0_HI"/>
+	<reg32 offset="0x0542" name="RBBM_PRIMCTR_1_LO"/>
+	<reg32 offset="0x0543" name="RBBM_PRIMCTR_1_HI"/>
+	<reg32 offset="0x0544" name="RBBM_PRIMCTR_2_LO"/>
+	<reg32 offset="0x0545" name="RBBM_PRIMCTR_2_HI"/>
+	<reg32 offset="0x0546" name="RBBM_PRIMCTR_3_LO"/>
+	<reg32 offset="0x0547" name="RBBM_PRIMCTR_3_HI"/>
+	<reg32 offset="0x0548" name="RBBM_PRIMCTR_4_LO"/>
+	<reg32 offset="0x0549" name="RBBM_PRIMCTR_4_HI"/>
+	<reg32 offset="0x054a" name="RBBM_PRIMCTR_5_LO"/>
+	<reg32 offset="0x054b" name="RBBM_PRIMCTR_5_HI"/>
+	<reg32 offset="0x054c" name="RBBM_PRIMCTR_6_LO"/>
+	<reg32 offset="0x054d" name="RBBM_PRIMCTR_6_HI"/>
+	<reg32 offset="0x054e" name="RBBM_PRIMCTR_7_LO"/>
+	<reg32 offset="0x054f" name="RBBM_PRIMCTR_7_HI"/>
+	<!-- PRIMCTR_8 appears to count number of primitives coming out of GS -->
+	<reg32 offset="0x0550" name="RBBM_PRIMCTR_8_LO"/>
+	<reg32 offset="0x0551" name="RBBM_PRIMCTR_8_HI"/>
+	<reg32 offset="0x0552" name="RBBM_PRIMCTR_9_LO"/>
+	<reg32 offset="0x0553" name="RBBM_PRIMCTR_9_HI"/>
+	<reg32 offset="0x0554" name="RBBM_PRIMCTR_10_LO"/>
+	<reg32 offset="0x0555" name="RBBM_PRIMCTR_10_HI"/>
+	
+
 	<reg32 offset="0xF400" name="RBBM_SECVID_TRUST_CNTL"/>
 	<reg32 offset="0xF800" name="RBBM_SECVID_TSB_TRUSTED_BASE_LO"/>
 	<reg32 offset="0xF801" name="RBBM_SECVID_TSB_TRUSTED_BASE_HI"/>
diff --git a/src/freedreno/registers/adreno_pm4.xml b/src/freedreno/registers/adreno_pm4.xml
index 23705a7cad9..f94e8e4e10b 100644
--- a/src/freedreno/registers/adreno_pm4.xml
+++ b/src/freedreno/registers/adreno_pm4.xml
@@ -16,6 +16,8 @@ xsi:schemaLocation="http://nouveau.freedesktop.org/ rules-ng.xsd">
 	<value name="VIZQUERY_END" value="8"/>
 	<value name="SC_WAIT_WC" value="9"/>
 	<value name="WRITE_PRIMITIVE_COUNTS" value="9" variants="A6XX"/>
+	<value name="START_PRIMITIVE_CTRS" value="11" variants="A6XX"/>
+	<value name="STOP_PRIMITIVE_CTRS" value="12" variants="A6XX"/>
 	<value name="RST_PIX_CNT" value="13"/>
 	<value name="RST_VTX_CNT" value="14"/>
 	<value name="TILE_FLUSH" value="15"/>
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_query.c b/src/gallium/drivers/freedreno/a6xx/fd6_query.c
index d950144bc7e..db249fc6ef7 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_query.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_query.c
@@ -256,6 +256,8 @@ struct PACKED fd6_primitives_sample {
 	struct {
 		uint64_t generated, emitted;
 	} start[4], stop[4], result;
+
+	uint64_t prim_start[16], prim_stop[16], prim_emitted;
 };
 
 
@@ -268,6 +270,17 @@ static void
 primitive_counts_resume(struct fd_acc_query *aq, struct fd_batch *batch)
 {
 	struct fd_ringbuffer *ring = batch->draw;
+	const unsigned count = 1;
+
+	fd_wfi(batch, ring);
+
+	OUT_PKT7(ring, CP_REG_TO_MEM, 3);
+	OUT_RING(ring, CP_REG_TO_MEM_0_64B |
+			CP_REG_TO_MEM_0_CNT(count - 1) |
+			CP_REG_TO_MEM_0_REG(REG_A6XX_RBBM_PRIMCTR_8_LO));
+	primitives_relocw(ring, aq, prim_start);
+
+	fd6_event_write(batch, ring, START_PRIMITIVE_CTRS, false);
 
 	OUT_PKT4(ring, REG_A6XX_VPC_SO_STREAM_COUNTS_LO, 2);
 	primitives_relocw(ring, aq, start[0]);
@@ -279,6 +292,25 @@ static void
 primitive_counts_pause(struct fd_acc_query *aq, struct fd_batch *batch)
 {
 	struct fd_ringbuffer *ring = batch->draw;
+	const unsigned count = 1;
+
+	fd_wfi(batch, ring);
+
+	/* snapshot the end values: */
+	OUT_PKT7(ring, CP_REG_TO_MEM, 3);
+	OUT_RING(ring, CP_REG_TO_MEM_0_64B |
+			CP_REG_TO_MEM_0_CNT(count - 1) |
+			CP_REG_TO_MEM_0_REG(REG_A6XX_RBBM_PRIMCTR_8_LO));
+	primitives_relocw(ring, aq, prim_stop);
+
+	/* result += stop - start: */
+	OUT_PKT7(ring, CP_MEM_TO_MEM, 9);
+	OUT_RING(ring, CP_MEM_TO_MEM_0_DOUBLE |
+			CP_MEM_TO_MEM_0_NEG_C | 0x40000000);
+	primitives_relocw(ring, aq, result.generated);
+	primitives_reloc(ring, aq, prim_emitted);
+	primitives_reloc(ring, aq, prim_stop);
+	primitives_reloc(ring, aq, prim_start);
 
 	OUT_PKT4(ring, REG_A6XX_VPC_SO_STREAM_COUNTS_LO, 2);
 	primitives_relocw(ring, aq, stop[0]);
@@ -304,6 +336,8 @@ primitive_counts_pause(struct fd_acc_query *aq, struct fd_batch *batch)
 	primitives_reloc(ring, aq, result.generated);
 	primitives_reloc(ring, aq, stop[aq->base.index].generated);
 	primitives_reloc(ring, aq, start[aq->base.index].generated);
+
+	fd6_event_write(batch, ring, STOP_PRIMITIVE_CTRS, false);
 }
 
 static void