radv: fix begin/end transform feedback with 0 counter buffers.

If the user gives 0 counterBuffers then the driver should still
enable transform feedback on all targets. This changes the
driver to always enable xfb, and use counter buffers where
one is defined for the target in question.

Fixes: b4eb029062 (radv: implement VK_EXT_transform_feedback)
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
(cherry picked from commit 677b496b6b)

src/amd/vulkan/radv_cmd_buffer.c

@@ -4743,13 +4743,15 @@ void radv_CmdBeginTransformFeedbackEXT(
 	struct radv_streamout_binding *sb = cmd_buffer->streamout_bindings;
 	struct radv_streamout_state *so = &cmd_buffer->state.streamout;
 	struct radeon_cmdbuf *cs = cmd_buffer->cs;
+	uint32_t i;
 
 	radv_flush_vgt_streamout(cmd_buffer);
 
 	assert(firstCounterBuffer + counterBufferCount <= MAX_SO_BUFFERS);
-	for (uint32_t i = firstCounterBuffer; i < counterBufferCount; i++) {
-		if (!(so->enabled_mask & (1 << i)))
-			continue;
+	for_each_bit(i, so->enabled_mask) {
+		int32_t counter_buffer_idx = i - firstCounterBuffer;
+		if (counter_buffer_idx >= 0 && counter_buffer_idx > counterBufferCount)
+			counter_buffer_idx = -1;
 
 		/* SI binds streamout buffers as shader resources.
 		 * VGT only counts primitives and tells the shader through
@@ -4759,12 +4761,12 @@ void radv_CmdBeginTransformFeedbackEXT(
 		radeon_emit(cs, sb[i].size >> 2);	/* BUFFER_SIZE (in DW) */
 		radeon_emit(cs, so->stride_in_dw[i]);			/* VTX_STRIDE (in DW) */
 
-		if (pCounterBuffers && pCounterBuffers[i]) {
+		if (counter_buffer_idx >= 0 && pCounterBuffers && pCounterBuffers[counter_buffer_idx]) {
 			/* The array of counter buffers is optional. */
-			RADV_FROM_HANDLE(radv_buffer, buffer, pCounterBuffers[i]);
+			RADV_FROM_HANDLE(radv_buffer, buffer, pCounterBuffers[counter_buffer_idx]);
 			uint64_t va = radv_buffer_get_va(buffer->bo);
 
-			va += buffer->offset + pCounterBufferOffsets[i];
+			va += buffer->offset + pCounterBufferOffsets[counter_buffer_idx];
 
 			/* Append */
 			radeon_emit(cs, PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0));
@@ -4803,20 +4805,22 @@ void radv_CmdEndTransformFeedbackEXT(
 	RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
 	struct radv_streamout_state *so = &cmd_buffer->state.streamout;
 	struct radeon_cmdbuf *cs = cmd_buffer->cs;
+	uint32_t i;
 
 	radv_flush_vgt_streamout(cmd_buffer);
 
 	assert(firstCounterBuffer + counterBufferCount <= MAX_SO_BUFFERS);
-	for (uint32_t i = firstCounterBuffer; i < counterBufferCount; i++) {
-		if (!(so->enabled_mask & (1 << i)))
-			continue;
+	for_each_bit(i, so->enabled_mask) {
+		int32_t counter_buffer_idx = i - firstCounterBuffer;
+		if (counter_buffer_idx >= 0 && counter_buffer_idx > counterBufferCount)
+			counter_buffer_idx = -1;
 
-		if (pCounterBuffers && pCounterBuffers[i]) {
+		if (counter_buffer_idx >= 0 && pCounterBuffers && pCounterBuffers[counter_buffer_idx]) {
 			/* The array of counters buffer is optional. */
-			RADV_FROM_HANDLE(radv_buffer, buffer, pCounterBuffers[i]);
+			RADV_FROM_HANDLE(radv_buffer, buffer, pCounterBuffers[counter_buffer_idx]);
 			uint64_t va = radv_buffer_get_va(buffer->bo);
 
-			va += buffer->offset + pCounterBufferOffsets[i];
+			va += buffer->offset + pCounterBufferOffsets[counter_buffer_idx];
 
 			radeon_emit(cs, PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0));
 			radeon_emit(cs, STRMOUT_SELECT_BUFFER(i) |