Mesa (staging/22.1): radv: fix handling divisor == 0 with dynamic vertex input state
GitLab Mirror
gitlab-mirror at kemper.freedesktop.org
Wed Apr 27 19:30:40 UTC 2022
Module: Mesa
Branch: staging/22.1
Commit: 2eba273314aa1f704da43a3e2c29f1fcfac2c81d
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=2eba273314aa1f704da43a3e2c29f1fcfac2c81d
Author: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Date: Mon Apr 18 14:46:09 2022 +0200
radv: fix handling divisor == 0 with dynamic vertex input state
When the divisor is 0, the compiler should generate a different VS
prolog instead of re-using a previous prolog that uses nontrivial
divisors. This is because divisor == 0 and divisor > 1 should use
a different path to guarantee that the index is correctly computed.
Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02 at gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16009>
(cherry picked from commit f525706e7725345e187f076118feb3937dfb8c43)
---
.pick_status.json | 2 +-
src/amd/vulkan/radv_cmd_buffer.c | 17 +++++++++++++----
src/amd/vulkan/radv_shader.h | 1 +
3 files changed, 15 insertions(+), 5 deletions(-)
diff --git a/.pick_status.json b/.pick_status.json
index ec16b277e8a..3597210baaf 100644
--- a/.pick_status.json
+++ b/.pick_status.json
@@ -256,7 +256,7 @@
"description": "radv: fix handling divisor == 0 with dynamic vertex input state",
"nominated": true,
"nomination_type": 0,
- "resolution": 0,
+ "resolution": 1,
"main_sha": null,
"because_sha": null
},
diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index 8ab0b5aaacc..1727340675a 100644
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -2753,8 +2753,9 @@ union vs_prolog_key_header {
uint32_t misaligned_mask : 1;
uint32_t post_shuffle : 1;
uint32_t nontrivial_divisors : 1;
+ uint32_t zero_divisors : 1;
/* We need this to ensure the padding is zero. It's useful even if it's unused. */
- uint32_t padding0 : 6;
+ uint32_t padding0 : 5;
};
uint32_t v;
};
@@ -2796,6 +2797,7 @@ lookup_vs_prolog(struct radv_cmd_buffer *cmd_buffer, struct radv_shader *vs_shad
uint32_t attribute_mask = BITFIELD_MASK(num_attributes);
uint32_t instance_rate_inputs = state->instance_rate_inputs & attribute_mask;
+ uint32_t zero_divisors = state->zero_divisors & attribute_mask;
*nontrivial_divisors = state->nontrivial_divisors & attribute_mask;
enum chip_class chip = device->physical_device->rad_info.chip_class;
const uint32_t misaligned_mask = chip == GFX6 || chip >= GFX10 ? cmd_buffer->state.vbo_misaligned_mask : 0;
@@ -2807,7 +2809,7 @@ lookup_vs_prolog(struct radv_cmd_buffer *cmd_buffer, struct radv_shader *vs_shad
!misaligned_mask && !state->alpha_adjust_lo && !state->alpha_adjust_hi) {
if (!instance_rate_inputs) {
prolog = device->simple_vs_prologs[num_attributes - 1];
- } else if (num_attributes <= 16 && !*nontrivial_divisors &&
+ } else if (num_attributes <= 16 && !*nontrivial_divisors && !zero_divisors &&
util_bitcount(instance_rate_inputs) ==
(util_last_bit(instance_rate_inputs) - ffs(instance_rate_inputs) + 1)) {
unsigned index = radv_instance_rate_prolog_index(num_attributes, instance_rate_inputs);
@@ -2818,7 +2820,7 @@ lookup_vs_prolog(struct radv_cmd_buffer *cmd_buffer, struct radv_shader *vs_shad
return prolog;
/* if we couldn't use a pre-compiled prolog, find one in the cache or create one */
- uint32_t key_words[16];
+ uint32_t key_words[17];
unsigned key_size = 1;
struct radv_vs_prolog_key key;
@@ -2847,6 +2849,10 @@ lookup_vs_prolog(struct radv_cmd_buffer *cmd_buffer, struct radv_shader *vs_shad
header.nontrivial_divisors = true;
key_words[key_size++] = *nontrivial_divisors;
}
+ if (zero_divisors) {
+ header.zero_divisors = true;
+ key_words[key_size++] = zero_divisors;
+ }
if (misaligned_mask) {
header.misaligned_mask = true;
key_words[key_size++] = misaligned_mask;
@@ -5603,8 +5609,11 @@ radv_CmdSetVertexInputEXT(VkCommandBuffer commandBuffer, uint32_t vertexBindingD
if (binding->inputRate == VK_VERTEX_INPUT_RATE_INSTANCE) {
state->instance_rate_inputs |= 1u << loc;
state->divisors[loc] = binding->divisor;
- if (binding->divisor != 1)
+ if (binding->divisor == 0) {
+ state->zero_divisors |= 1u << loc;
+ } else if (binding->divisor > 1) {
state->nontrivial_divisors |= 1u << loc;
+ }
}
cmd_buffer->vertex_bindings[attrib->binding].stride = binding->stride;
state->offsets[loc] = attrib->offset;
diff --git a/src/amd/vulkan/radv_shader.h b/src/amd/vulkan/radv_shader.h
index c3a5ecc01c2..862d9ba6482 100644
--- a/src/amd/vulkan/radv_shader.h
+++ b/src/amd/vulkan/radv_shader.h
@@ -372,6 +372,7 @@ struct radv_vs_input_state {
uint32_t instance_rate_inputs;
uint32_t nontrivial_divisors;
+ uint32_t zero_divisors;
uint32_t post_shuffle;
/* Having two separate fields instead of a single uint64_t makes it easier to remove attributes
* using bitwise arithmetic.
More information about the mesa-commit
mailing list