[PATCH 0/2] MR448: hlsl: Properly handle allocation ranges that aren't aligned to a whole register.

[PATCH 1/2] vkd3d-shader/hlsl: Check that a partial register's mask is also available in is_range_available().

[Zebediah Figura]

From: Zebediah Figura

---
 libs/vkd3d-shader/hlsl_codegen.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
index 6eac5d490..456c393e3 100644
--- a/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d-shader/hlsl_codegen.c
@@ -3556,13 +3556,19 @@ static struct hlsl_reg allocate_register(struct hlsl_ctx *ctx, struct register_a
 static bool is_range_available(const struct register_allocator *allocator,
         unsigned int first_write, unsigned int last_read, uint32_t reg_idx, unsigned int reg_size)
 {
+    unsigned int last_reg_mask = (1u << (reg_size % 4)) - 1;
+    unsigned int writemask;
     uint32_t i;

for (i = 0; i < (reg_size / 4); ++i)
     {
-        if (get_available_writemask(allocator, first_write, last_read, reg_idx + i) != VKD3DSP_WRITEMASK_ALL)
+        writemask = get_available_writemask(allocator, first_write, last_read, reg_idx + i);
+        if (writemask != VKD3DSP_WRITEMASK_ALL)
             return false;
     }
+    writemask = get_available_writemask(allocator, first_write, last_read, reg_idx + (reg_size / 4));
+    if ((writemask & last_reg_mask) != last_reg_mask)
+        return false;
     return true;
 }

[PATCH 2/2] vkd3d-shader/hlsl: Record partial allocations in allocate_range().

[Zebediah Figura]

From: Zebediah Figura

---
 libs/vkd3d-shader/hlsl_codegen.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
index 456c393e3..76a010fb3 100644
--- a/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d-shader/hlsl_codegen.c
@@ -3587,6 +3587,8 @@ static struct hlsl_reg allocate_range(struct hlsl_ctx *ctx, struct register_allo

for (i = 0; i < reg_size / 4; ++i)
         record_allocation(ctx, allocator, reg_idx + i, VKD3DSP_WRITEMASK_ALL, first_write, last_read);
+    if (reg_size % 4)
+        record_allocation(ctx, allocator, reg_idx + (reg_size / 4), (1u << (reg_size % 4)) - 1, first_write, last_read);

ret.id = reg_idx;
     ret.allocation_size = align(reg_size, 4) / 4;

[PATCH 0/2] MR448: hlsl: Properly handle allocation ranges that aren't aligned to a whole register. - approved

[Giovanni Mascellani (＠giomasce)]

This merge request was approved by Giovanni Mascellani.

[PATCH 0/2] MR448: hlsl: Properly handle allocation ranges that aren't aligned to a whole register. - approved

[Henri Verbeet (＠hverbeet)]

This merge request was approved by Henri Verbeet.