summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorMarek Olšák <[email protected]>2017-10-08 18:20:38 +0200
committerMarek Olšák <[email protected]>2017-10-17 22:03:03 +0200
commitcdb21dfffa1300b1032619bc44da99c970fd3f56 (patch)
tree092c74c794efdaf13204d5bb705d895f4425a2ef
parentee0e1a47ce11c9199bb4a132edd5f622051f989c (diff)
radeonsi: handle 64-bit loads earlier in fetch_constant
Reviewed-by: Nicolai Hähnle <[email protected]>
-rw-r--r--src/gallium/drivers/radeonsi/si_shader.c26
1 files changed, 10 insertions, 16 deletions
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 62a056dce43..ff372ae0531 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -1977,7 +1977,6 @@ static LLVMValueRef fetch_constant(
unsigned buf, idx;
LLVMValueRef addr, bufp;
- LLVMValueRef result;
if (swizzle == LP_CHAN_ALL) {
unsigned chan;
@@ -1988,6 +1987,15 @@ static LLVMValueRef fetch_constant(
return lp_build_gather_values(&ctx->gallivm, values, 4);
}
+ /* Split 64-bit loads. */
+ if (tgsi_type_is_64bit(type)) {
+ LLVMValueRef lo, hi;
+
+ lo = fetch_constant(bld_base, reg, TGSI_TYPE_UNSIGNED, swizzle);
+ hi = fetch_constant(bld_base, reg, TGSI_TYPE_UNSIGNED, swizzle + 1);
+ return si_llvm_emit_fetch_64bit(bld_base, type, lo, hi);
+ }
+
assert(reg->Register.Dimension);
buf = reg->Dimension.Index;
idx = reg->Register.Index * 4 + swizzle;
@@ -2010,21 +2018,7 @@ static LLVMValueRef fetch_constant(
addr = LLVMConstInt(ctx->i32, idx * 4, 0);
}
- result = buffer_load_const(ctx, bufp, addr);
-
- if (!tgsi_type_is_64bit(type))
- result = bitcast(bld_base, type, result);
- else {
- LLVMValueRef addr2, result2;
-
- addr2 = lp_build_add(&bld_base->uint_bld, addr,
- LLVMConstInt(ctx->i32, 4, 0));
- result2 = buffer_load_const(ctx, bufp, addr2);
-
- result = si_llvm_emit_fetch_64bit(bld_base, type,
- result, result2);
- }
- return result;
+ return bitcast(bld_base, type, buffer_load_const(ctx, bufp, addr));
}
/* Upper 16 bits must be zero. */