[llvm-branch-commits] [llvm] AMDGPU: Include callees in backend_stack_size PAL metadata (PR #234522)
Diana Picus via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Mon Nov 16 05:03:46 PST 2026
https://github.com/rovka updated https://github.com/llvm/llvm-project/pull/234522
>From 2652a6d29bdd457c8ce34b3de6a76ad28596bdab Mon Sep 17 00:00:00 2001
From: Diana Picus <diana-magda.picus at amd.com>
Date: Wed, 14 Oct 2026 11:08:10 +0200
Subject: [PATCH 1/4] AMDGPU: Include callees in backend_stack_size PAL
metadata
The stack size reported in the PAL metadata of a chain function should
include the stack sizes of any callees, since the consumers don't have
access to the call graph (this is especially important for whole wave
functions, because we don't even emit any PAL metadata for them). The
consumers are generally expected to allocate scratch based on the
maximum `.backend_stack_size` emitted for the functions in a pipeline.
They will also have to deal with recursion and indirect calls.
The `.stack_frame_size_in_bytes` metadata remains unchanged, i.e. it
represents just the stack size of the current function (without its
callees).
Assisted-by: AI
---
llvm/docs/AMDGPUUsage.rst | 7 ++-
llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp | 3 +-
.../Target/AMDGPU/Utils/AMDGPUPALMetadata.cpp | 12 +++-
.../Target/AMDGPU/Utils/AMDGPUPALMetadata.h | 10 +++-
llvm/test/CodeGen/AMDGPU/amdpal-callable.ll | 4 +-
.../CodeGen/AMDGPU/amdpal-chain-metadata.ll | 60 +++++++++++++++++--
.../AMDGPU/pal-metadata-3.0-callable-dvgpr.ll | 4 +-
.../AMDGPU/pal-metadata-3.0-callable.ll | 4 +-
8 files changed, 87 insertions(+), 17 deletions(-)
diff --git a/llvm/docs/AMDGPUUsage.rst b/llvm/docs/AMDGPUUsage.rst
index 3de94d5617ac7b..53cbb657ef9811 100644
--- a/llvm/docs/AMDGPUUsage.rst
+++ b/llvm/docs/AMDGPUUsage.rst
@@ -20999,7 +20999,12 @@ within a map that has been added by the same *vendor-name*.
".lds_size" integer Size in bytes of LDS memory.
".vgpr_count" integer Number of VGPRs used by the shader.
".sgpr_count" integer Number of SGPRs used by the shader.
- ".stack_frame_size_in_bytes" integer Amount of stack size used by the shader.
+ ".stack_frame_size_in_bytes" integer Amount of stack size used by the shader itself, not including
+ the stack used by any of its callees.
+ ".backend_stack_size" integer Amount of stack size used by the shader, including the stack
+ used by its known callees. Chain callees are excluded, since they
+ reuse the caller's stack frame. Indirect calls and recursion do
+ not contribute.
".shader_subtype" string Shader subtype/kind. Values include:
- "Unknown"
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
index f2761dc9755b76..9b326a12521c36 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -1815,7 +1815,8 @@ void AMDGPUAsmPrinter::emitPALFunctionMetadata(const MachineFunction &MF) {
auto *MD = getTargetStreamer()->getPALMetadata();
const MachineFrameInfo &MFI = MF.getFrameInfo();
StringRef FnName = MF.getFunction().getName();
- MD->setFunctionScratchSize(FnName, MFI.getStackSize());
+ MD->setFunctionStackFrameSize(FnName, MFI.getStackSize());
+ MD->setFunctionBackendStackSize(FnName, CurrentProgramInfo.ScratchSize);
const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
MCContext &Ctx = MF.getContext();
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.cpp b/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.cpp
index 14a9a217c01cc0..33d0c5e7cbc7a9 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.cpp
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.cpp
@@ -366,11 +366,17 @@ void AMDGPUPALMetadata::setScratchSize(unsigned CC, const MCExpr *Val,
setHwStage(CC, ".scratch_memory_size", msgpack::Type::UInt, Val);
}
-// Set the stack frame size of a function in the metadata.
-void AMDGPUPALMetadata::setFunctionScratchSize(StringRef FnName, unsigned Val) {
+void AMDGPUPALMetadata::setFunctionStackFrameSize(StringRef FnName,
+ unsigned Val) {
auto Node = getShaderFunction(FnName);
Node[".stack_frame_size_in_bytes"] = MsgPackDoc.getNode(Val);
- Node[".backend_stack_size"] = MsgPackDoc.getNode(Val);
+}
+
+void AMDGPUPALMetadata::setFunctionBackendStackSize(StringRef FnName,
+ const MCExpr *Val) {
+ msgpack::MapDocNode Node = getShaderFunction(FnName);
+ DelayedExprs.assignDocNode(Node[".backend_stack_size"], msgpack::Type::UInt,
+ Val);
}
// Set the amount of LDS used in bytes in the metadata.
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.h b/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.h
index 6669cf387c5b56..e619d140fd8395 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.h
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.h
@@ -103,8 +103,14 @@ class AMDGPUPALMetadata {
void setScratchSize(unsigned CC, unsigned Val);
void setScratchSize(unsigned CC, const MCExpr *Val, MCContext &Ctx);
- // Set the stack frame size of a function in the metadata.
- void setFunctionScratchSize(StringRef FnName, unsigned Val);
+ // Set the stack frame size of a function in the metadata. This only covers
+ // the frame of the function itself, not the frames of any of its callees.
+ void setFunctionStackFrameSize(StringRef FnName, unsigned Val);
+
+ // Set the backend stack size metadata, including the frames of its known
+ // calles. Recursion and indirect/external calls must be handled by the
+ // consumer.
+ void setFunctionBackendStackSize(StringRef FnName, const MCExpr *Val);
// Set the amount of LDS used in bytes in the metadata. This is an optional
// advisory record for logging etc; wave dispatch actually uses the rsrc1
diff --git a/llvm/test/CodeGen/AMDGPU/amdpal-callable.ll b/llvm/test/CodeGen/AMDGPU/amdpal-callable.ll
index 20e9842d5e62b8..14bb9dc9da29e2 100644
--- a/llvm/test/CodeGen/AMDGPU/amdpal-callable.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdpal-callable.ll
@@ -171,7 +171,7 @@ attributes #0 = { nounwind }
; GCN-NEXT: .stack_frame_size_in_bytes: 0{{$}}
; GCN-NEXT: .vgpr_count: 0x1{{$}}
; GCN-NEXT: no_stack_call:
-; GCN-NEXT: .backend_stack_size: 0x10{{$}}
+; GCN-NEXT: .backend_stack_size: 0x24{{$}}
; GCN-NEXT: .lds_size: 0{{$}}
; GCN-NEXT: .sgpr_count: 0x25{{$}}
; GCN-NEXT: .stack_frame_size_in_bytes: 0x10{{$}}
@@ -216,7 +216,7 @@ attributes #0 = { nounwind }
; GCN-NEXT: .stack_frame_size_in_bytes: 0x14{{$}}
; GCN-NEXT: .vgpr_count: 0x2{{$}}
; GCN-NEXT: simple_stack_call:
-; GCN-NEXT: .backend_stack_size: 0x20{{$}}
+; GCN-NEXT: .backend_stack_size: 0x34{{$}}
; GCN-NEXT: .lds_size: 0{{$}}
; GCN-NEXT: .sgpr_count: 0x25{{$}}
; GCN-NEXT: .stack_frame_size_in_bytes: 0x20{{$}}
diff --git a/llvm/test/CodeGen/AMDGPU/amdpal-chain-metadata.ll b/llvm/test/CodeGen/AMDGPU/amdpal-chain-metadata.ll
index d972dd7884902e..9d8da278ac79c8 100644
--- a/llvm/test/CodeGen/AMDGPU/amdpal-chain-metadata.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdpal-chain-metadata.ll
@@ -1,4 +1,4 @@
-; RUN: llc -mtriple=amdgpu11.00--amdpal <%s | FileCheck %s
+; RUN: llc -mtriple=amdgpu11.00--amdpal < %s | FileCheck %s
; CHECK-LABEL: {{^}}amdgpu_cs_chain_func:
; CHECK: .set .Lamdgpu_cs_chain_func.private_seg_size, 16{{$}}
@@ -28,6 +28,19 @@
; CHECK: .set .Lchain_indirect_and_wwf.private_seg_size, 64+max(.Lwwf_alloca.private_seg_size){{$}}
; CHECK: .set .Lchain_indirect_and_wwf.has_indirect_call, 1{{$}}
+; CHECK-LABEL: {{^}}chain_indirect_gfx_and_wwf:
+; CHECK: .set .Lchain_indirect_gfx_and_wwf.private_seg_size, 64+max(.Lwwf_alloca.private_seg_size){{$}}
+; CHECK: .set .Lchain_indirect_gfx_and_wwf.has_indirect_call, 1{{$}}
+
+; CHECK-LABEL: {{^}}gfx_stack_only:
+; CHECK: .set .Lgfx_stack_only.private_seg_size, 20{{$}}
+
+; CHECK-LABEL: {{^}}wwf_callee_stack_only:
+; CHECK: .set .Lwwf_callee_stack_only.private_seg_size, 16+max(.Lgfx_stack_only.private_seg_size){{$}}
+
+; CHECK-LABEL: {{^}}chain_own_and_callee_stack:
+; CHECK: .set .Lchain_own_and_callee_stack.private_seg_size, 32+max(.Lwwf_callee_stack_only.private_seg_size){{$}}
+
; CHECK-LABEL: .amdgpu_pal_metadata
; CHECK-NEXT: ---
; CHECK-NEXT: amdpal.pipelines:
@@ -41,10 +54,10 @@
; CHECK: .backend_stack_size: 0x10{{$}}
; CHECK: .stack_frame_size_in_bytes: 0x10{{$}}
; CHECK-LABEL: chain_and_gfx:
-; CHECK: .backend_stack_size: 0x20{{$}}
+; CHECK: .backend_stack_size: 0x3c{{$}}
; CHECK: .stack_frame_size_in_bytes: 0x20{{$}}
; CHECK-LABEL: chain_and_wwf:
-; CHECK: .backend_stack_size: 0x50{{$}}
+; CHECK: .backend_stack_size: 0x6c{{$}}
; CHECK: .stack_frame_size_in_bytes: 0x50{{$}}
; CHECK-LABEL: chain_callee_larger_stack:
; CHECK: .backend_stack_size: 0x8{{$}}
@@ -53,11 +66,20 @@
; CHECK: .backend_stack_size: 0x24{{$}}
; CHECK: .stack_frame_size_in_bytes: 0x24{{$}}
; CHECK-LABEL: chain_indirect_and_wwf:
-; CHECK: .backend_stack_size: 0x40{{$}}
+; CHECK: .backend_stack_size: 0x5c{{$}}
+; CHECK: .stack_frame_size_in_bytes: 0x40{{$}}
+; CHECK-LABEL: chain_indirect_gfx_and_wwf:
+; CHECK: .backend_stack_size: 0x5c{{$}}
; CHECK: .stack_frame_size_in_bytes: 0x40{{$}}
+; CHECK-LABEL: chain_own_and_callee_stack:
+; CHECK: .backend_stack_size: 0x44{{$}}
+; CHECK: .stack_frame_size_in_bytes: 0x20{{$}}
; CHECK-LABEL: gfx_alloca:
; CHECK: .backend_stack_size: 0x1c{{$}}
; CHECK: .stack_frame_size_in_bytes: 0x1c{{$}}
+; CHECK-LABEL: gfx_stack_only:
+; CHECK: .backend_stack_size: 0x14{{$}}
+; CHECK: .stack_frame_size_in_bytes: 0x14{{$}}
; CHECK:amdpal.version:
; CHECK-NEXT: - 0x3
; CHECK-NEXT: - 0
@@ -137,11 +159,41 @@ define amdgpu_cs_chain void @chain_indirect_and_wwf(ptr inreg %callee, <40 x i32
unreachable
}
+define amdgpu_cs_chain void @chain_indirect_gfx_and_wwf(ptr inreg %callee) {
+.entry:
+ %a = alloca [12 x i32], addrspace(5)
+ store i32 11, ptr addrspace(5) %a
+ call void (ptr, ...) @llvm.amdgcn.call.whole.wave(ptr @wwf_alloca, ptr addrspace(5) %a)
+ call amdgpu_gfx void %callee()
+ ret void
+}
+
define amdgpu_cs_chain_preserve void @amdgpu_cs_chain_preserve_func() {
.entry:
ret void
}
+; Stack sizes accumulate transitively through non-chain callees, across calling
+; conventions.
+define amdgpu_gfx void @gfx_stack_only() {
+ %a = alloca [4 x i32], addrspace(5)
+ store volatile i32 12, ptr addrspace(5) %a
+ ret void
+}
+
+define amdgpu_gfx_whole_wave void @wwf_callee_stack_only(i1 %active) {
+ call amdgpu_gfx void @gfx_stack_only()
+ ret void
+}
+
+define amdgpu_cs_chain void @chain_own_and_callee_stack() {
+.entry:
+ %a = alloca [4 x i32], addrspace(5)
+ store volatile i32 13, ptr addrspace(5) %a
+ call void (ptr, ...) @llvm.amdgcn.call.whole.wave(ptr @wwf_callee_stack_only)
+ ret void
+}
+
declare void @llvm.amdgcn.call.whole.wave(ptr, ...)
declare void @llvm.amdgcn.cs.chain.p0.i32.s.v40i32(ptr, i32, {}, <40 x i32>, i32 immarg, ...)
diff --git a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable-dvgpr.ll b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable-dvgpr.ll
index 1c570d6d21358a..95bea92f331e96 100644
--- a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable-dvgpr.ll
+++ b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable-dvgpr.ll
@@ -91,7 +91,7 @@
; CHECK-NEXT: .stack_frame_size_in_bytes: 0
; CHECK-NEXT: .vgpr_count: 0x1
; CHECK-NEXT: no_stack_call:
-; CHECK-NEXT: .backend_stack_size: 0x10
+; CHECK-NEXT: .backend_stack_size: 0x24
; CHECK-NEXT: .lds_size: 0
; CHECK-NEXT: .sgpr_count: 0x22
; CHECK-NEXT: .stack_frame_size_in_bytes: 0x10
@@ -133,7 +133,7 @@
; CHECK-NEXT: .stack_frame_size_in_bytes: 0x14
; CHECK-NEXT: .vgpr_count: 0x2
; CHECK-NEXT: simple_stack_call:
-; CHECK-NEXT: .backend_stack_size: 0x20
+; CHECK-NEXT: .backend_stack_size: 0x34
; CHECK-NEXT: .lds_size: 0
; CHECK-NEXT: .sgpr_count: 0x22
; CHECK-NEXT: .stack_frame_size_in_bytes: 0x20
diff --git a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable.ll b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable.ll
index ca0941e8cebc84..9f86587cd7bfef 100644
--- a/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable.ll
+++ b/llvm/test/CodeGen/AMDGPU/pal-metadata-3.0-callable.ll
@@ -92,7 +92,7 @@
; CHECK-NEXT: .stack_frame_size_in_bytes: 0
; CHECK-NEXT: .vgpr_count: 0x1
; CHECK-NEXT: no_stack_call:
-; CHECK-NEXT: .backend_stack_size: 0x10
+; CHECK-NEXT: .backend_stack_size: 0x24
; CHECK-NEXT: .lds_size: 0
; CHECK-NEXT: .sgpr_count: 0x22
; CHECK-NEXT: .stack_frame_size_in_bytes: 0x10
@@ -137,7 +137,7 @@
; CHECK-NEXT: .stack_frame_size_in_bytes: 0x14
; CHECK-NEXT: .vgpr_count: 0x2
; CHECK-NEXT: simple_stack_call:
-; CHECK-NEXT: .backend_stack_size: 0x20
+; CHECK-NEXT: .backend_stack_size: 0x34
; CHECK-NEXT: .lds_size: 0
; CHECK-NEXT: .sgpr_count: 0x22
; CHECK-NEXT: .stack_frame_size_in_bytes: 0x20
>From ae35701e0d2e0f81a653d8ad21938a06144b67d7 Mon Sep 17 00:00:00 2001
From: Diana Picus <diana-magda.picus at amd.com>
Date: Wed, 28 Oct 2026 10:41:23 +0100
Subject: [PATCH 2/4] Move docs to md instead of rst
---
llvm/docs/AMDGPUUsage.md | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/llvm/docs/AMDGPUUsage.md b/llvm/docs/AMDGPUUsage.md
index 3b02e0e92d5eb8..c6963d32a9b6cf 100644
--- a/llvm/docs/AMDGPUUsage.md
+++ b/llvm/docs/AMDGPUUsage.md
@@ -16667,7 +16667,12 @@ within a map that has been added by the same *vendor-name*.
- Number of SGPRs used by the shader.
* - ".stack_frame_size_in_bytes"
- integer
- - Amount of stack size used by the shader.
+ - Amount of stack size used by the shader itself, not including the stack used by any of its callees.
+* - ".backend_stack_size"
+ - integer
+ - Amount of stack size used by the shader, including the stack used by its known callees.
+ Chain callees are excluded, since they reuse the caller's stack frame.
+ Indirect calls and recursion do not contribute.
* - ".shader_subtype"
- string
- Shader subtype/kind.
>From 1961971ee84f0963d59e04a59663b5f69afdb9ab Mon Sep 17 00:00:00 2001
From: Diana Picus <diana-magda.picus at amd.com>
Date: Wed, 28 Oct 2026 12:21:50 +0100
Subject: [PATCH 3/4] More tests and minor fixes
---
.../Target/AMDGPU/Utils/AMDGPUPALMetadata.h | 2 +-
.../CodeGen/AMDGPU/amdpal-chain-metadata.ll | 18 +++++++++++++++++-
2 files changed, 18 insertions(+), 2 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.h b/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.h
index e619d140fd8395..6e64df993f8ff2 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.h
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUPALMetadata.h
@@ -108,7 +108,7 @@ class AMDGPUPALMetadata {
void setFunctionStackFrameSize(StringRef FnName, unsigned Val);
// Set the backend stack size metadata, including the frames of its known
- // calles. Recursion and indirect/external calls must be handled by the
+ // callees. Recursion and indirect/external calls must be handled by the
// consumer.
void setFunctionBackendStackSize(StringRef FnName, const MCExpr *Val);
diff --git a/llvm/test/CodeGen/AMDGPU/amdpal-chain-metadata.ll b/llvm/test/CodeGen/AMDGPU/amdpal-chain-metadata.ll
index 37e438c0e17838..ec25c56f203a08 100644
--- a/llvm/test/CodeGen/AMDGPU/amdpal-chain-metadata.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdpal-chain-metadata.ll
@@ -26,7 +26,6 @@
; CHECK-LABEL: {{^}}cs_to_chain_and_wwf:
; CHECK: .set .Lcs_to_chain_and_wwf.private_seg_size, 12{{$}}
-; Indirect chain calls don't contribute to the stack size.
; Indirect chain calls don't contribute to the stack size.
; CHECK-LABEL: {{^}}chain_indirect_chain_and_wwf:
; CHECK: .set .Lchain_indirect_chain_and_wwf.private_seg_size, 64+max(.Lwwf_callee_alloca.private_seg_size){{$}}
@@ -54,6 +53,10 @@
; EXTSTACK0: .set .Lchain_own_and_callee_stack.private_seg_size, 32+max(.Lwwf_no_local_stack_to_gfx_with_stack.private_seg_size){{$}}
; EXTSTACK48: .set .Lchain_own_and_callee_stack.private_seg_size, 32+max(48, .Lwwf_no_local_stack_to_gfx_with_stack.private_seg_size){{$}}
+; CHECK-LABEL: {{^}}chain_realign_stack:
+; EXTSTACK0: .set .Lchain_realign_stack.private_seg_size, 256+max(.Lgfx_stack_only.private_seg_size){{$}}
+; EXTSTACK48: .set .Lchain_realign_stack.private_seg_size, 256+max(48, .Lgfx_stack_only.private_seg_size){{$}}
+
; CHECK-LABEL: .amdgpu_pal_metadata
; CHECK-NEXT: ---
; CHECK-NEXT: amdpal.pipelines:
@@ -94,6 +97,11 @@
; EXTSTACK0: .backend_stack_size: 0x44{{$}}
; EXTSTACK48: .backend_stack_size: 0x60{{$}}
; CHECK: .stack_frame_size_in_bytes: 0x20{{$}}
+; CHECK-LABEL: chain_realign_stack:
+; EXTSTACK0: .backend_stack_size: 0x114{{$}}
+; EXTSTACK48: .backend_stack_size: 0x130{{$}}
+; FIXME: stack_frame_size_in_bytes would do well to include the realignment padding too...
+; CHECK: .stack_frame_size_in_bytes: 0x80{{$}}
; CHECK-LABEL: gfx_alloca:
; CHECK: .backend_stack_size: 0x1c{{$}}
; CHECK: .stack_frame_size_in_bytes: 0x1c{{$}}
@@ -221,6 +229,14 @@ define amdgpu_cs_chain void @chain_own_and_callee_stack() {
ret void
}
+define amdgpu_cs_chain void @chain_realign_stack() {
+.entry:
+ %a = alloca i32, align 128, addrspace(5)
+ store volatile i32 13, ptr addrspace(5) %a
+ call amdgpu_gfx void @gfx_stack_only()
+ ret void
+}
+
declare amdgpu_gfx void @use(...)
!amdgpu.pal.metadata.msgpack = !{!0}
>From 421fd90f6449fdbe11057fb99b4c62c02d22fbbe Mon Sep 17 00:00:00 2001
From: Diana Picus <diana-magda.picus at amd.com>
Date: Mon, 16 Nov 2026 14:02:42 +0100
Subject: [PATCH 4/4] in bytes
---
llvm/docs/AMDGPUUsage.md | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/docs/AMDGPUUsage.md b/llvm/docs/AMDGPUUsage.md
index 198ce3c7b26d1c..f7158adce27d05 100644
--- a/llvm/docs/AMDGPUUsage.md
+++ b/llvm/docs/AMDGPUUsage.md
@@ -16715,7 +16715,7 @@ within a map that has been added by the same *vendor-name*.
- Amount of stack size used by the shader itself, not including the stack used by any of its callees.
* - ".backend_stack_size"
- integer
- - Amount of stack size used by the shader, including the stack used by its known callees.
+ - Amount of stack size in bytes used by the shader, including the stack used by its known callees.
Chain callees are excluded, since they reuse the caller's stack frame.
Indirect calls and recursion do not contribute.
* - ".shader_subtype"
More information about the llvm-branch-commits
mailing list