From 7f10fe08d444aa9e9104387cb4cb2f0bc0ab1a12 Mon Sep 17 00:00:00 2001
From: Hans-Kristian Arntzen <post@arntzen-software.no>
Date: Fri, 4 Apr 2025 17:06:02 +0200
Subject: [PATCH] GLSL: Fix coopmat function parameter typing.

---
 .../cooperative-matrix.vk.nocompat.comp.vk    | 118 +++++++-------
 .../cooperative-matrix.vk.nocompat.comp.vk    | 152 ++++++++++--------
 .../comp/cooperative-matrix.vk.nocompat.comp  |  14 ++
 spirv_glsl.cpp                                |  16 +-
 4 files changed, 168 insertions(+), 132 deletions(-)

diff --git a/reference/opt/shaders/comp/cooperative-matrix.vk.nocompat.comp.vk b/reference/opt/shaders/comp/cooperative-matrix.vk.nocompat.comp.vk
index c730aed24..095573980 100644
--- a/reference/opt/shaders/comp/cooperative-matrix.vk.nocompat.comp.vk
+++ b/reference/opt/shaders/comp/cooperative-matrix.vk.nocompat.comp.vk
@@ -30,68 +30,68 @@ shared uint blah[512];
 
 void main()
 {
-    uint _514 = 256u * gl_WorkGroupID.x;
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _519;
-    coopMatLoad(_519, ssbo32.data, _514, 16u, Layout);
-    uint _523 = 512u * gl_WorkGroupID.x;
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _528;
-    coopMatLoad(_528, ssbo16.data, _523, 32u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _537;
-    coopMatLoad(_537, ssbo32.data, _514, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _546;
-    coopMatLoad(_546, ssbo16.data, _523, 32u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _555;
-    coopMatLoad(_555, ssbo32.data, _514, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _564;
-    coopMatLoad(_564, ssbo16.data, _523, 32u, Layout);
-    uint _578 = 128u * gl_WorkGroupID.x;
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> _583;
-    coopMatLoad(_583, ssbo32.data, _578, 8u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> _592;
-    coopMatLoad(_592, ssbo16.data, _514, 16u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> _601;
-    coopMatLoad(_601, ssbo32.data, _578, 8u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> _610;
-    coopMatLoad(_610, ssbo16.data, _514, 16u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> _619;
-    coopMatLoad(_619, ssbo32.data, _578, 8u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> _628;
-    coopMatLoad(_628, ssbo16.data, _514, 16u, Layout);
-    coopMatStore(coopmat<float, Scope, Rows, Columns, gl_MatrixUseA>(100.0), ssbo32.data, _578, 0u, Layout);
-    coopMatStore(coopmat<uint, Scope, Rows, Columns, gl_MatrixUseA>(100u), ssbo32.data, _578, 0u, Layout);
-    coopMatStore(coopmat<int, Scope, Rows, Columns, gl_MatrixUseA>(-100), ssbo32.data, _578, 0u, Layout);
-    coopMatStore(coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA>(float16_t(100.0)), ssbo32.data, _578, 0u, Layout);
-    coopMatStore(coopmat<int16_t, Scope, Rows, Columns, gl_MatrixUseA>(-100s), ssbo32.data, _578, 0u, Layout);
-    coopMatStore(coopmat<uint16_t, Scope, Rows, Columns, gl_MatrixUseA>(100us), ssbo32.data, _578, 0u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _726 = coopmat<float, Scope, Rows, Columns, gl_MatrixUseA>(100.0);
-    for (int _884 = 0; _884 < int(uint(coopmat<float, Scope, Rows, Columns, gl_MatrixUseA>(0).length())); )
+    uint _549 = 256u * gl_WorkGroupID.x;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _554;
+    coopMatLoad(_554, ssbo32.data, _549, 16u, Layout);
+    uint _558 = 512u * gl_WorkGroupID.x;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _563;
+    coopMatLoad(_563, ssbo16.data, _558, 32u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _572;
+    coopMatLoad(_572, ssbo32.data, _549, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _581;
+    coopMatLoad(_581, ssbo16.data, _558, 32u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _590;
+    coopMatLoad(_590, ssbo32.data, _549, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _599;
+    coopMatLoad(_599, ssbo16.data, _558, 32u, Layout);
+    uint _613 = 128u * gl_WorkGroupID.x;
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> _618;
+    coopMatLoad(_618, ssbo32.data, _613, 8u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> _627;
+    coopMatLoad(_627, ssbo16.data, _549, 16u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> _636;
+    coopMatLoad(_636, ssbo32.data, _613, 8u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> _645;
+    coopMatLoad(_645, ssbo16.data, _549, 16u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> _654;
+    coopMatLoad(_654, ssbo32.data, _613, 8u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> _663;
+    coopMatLoad(_663, ssbo16.data, _549, 16u, Layout);
+    coopMatStore(coopmat<float, Scope, Rows, Columns, gl_MatrixUseA>(100.0), ssbo32.data, _613, 0u, Layout);
+    coopMatStore(coopmat<uint, Scope, Rows, Columns, gl_MatrixUseA>(100u), ssbo32.data, _613, 0u, Layout);
+    coopMatStore(coopmat<int, Scope, Rows, Columns, gl_MatrixUseA>(-100), ssbo32.data, _613, 0u, Layout);
+    coopMatStore(coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA>(float16_t(100.0)), ssbo32.data, _613, 0u, Layout);
+    coopMatStore(coopmat<int16_t, Scope, Rows, Columns, gl_MatrixUseA>(-100s), ssbo32.data, _613, 0u, Layout);
+    coopMatStore(coopmat<uint16_t, Scope, Rows, Columns, gl_MatrixUseA>(100us), ssbo32.data, _613, 0u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _761 = coopmat<float, Scope, Rows, Columns, gl_MatrixUseA>(100.0);
+    for (int _944 = 0; _944 < int(uint(coopmat<float, Scope, Rows, Columns, gl_MatrixUseA>(0).length())); )
     {
-        _726[_884] += 50.0;
-        _884++;
+        _761[_944] += 50.0;
+        _944++;
         continue;
     }
-    coopMatStore(_726, ssbo32.data, 0u, 16u, Layout);
+    coopMatStore(_761, ssbo32.data, 0u, 16u, Layout);
     coopMatStore(coopmat<uint, Scope, Rows, Columns, gl_MatrixUseA>(gl_WorkGroupID.x), ssbo32.data, 0u, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _768;
-    coopMatLoad(_768, ssbo32.data, _514, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _777;
-    coopMatLoad(_777, ssbo32.data, _514, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _786;
-    coopMatLoad(_786, ssbo32.data, _514, 16u, Layout);
-    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseA> _814;
-    coopMatLoad(_814, ssbo32.data, _514, 16u, Layout);
-    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseB> _823;
-    coopMatLoad(_823, ssbo32.data, _514, 16u, Layout);
-    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseAccumulator> _832;
-    coopMatLoad(_832, ssbo32.data, _514, 16u, Layout);
-    coopmat<int, Scope, Rows, Columns, gl_MatrixUseA> _852;
-    coopMatLoad(_852, ssbo32.data, _514, 16u, Layout);
-    coopmat<int, Scope, Rows, Columns, gl_MatrixUseB> _861;
-    coopMatLoad(_861, ssbo32.data, _514, 16u, Layout);
-    coopmat<int, Scope, Rows, Columns, gl_MatrixUseAccumulator> _870;
-    coopMatLoad(_870, ssbo32.data, _514, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _880;
-    coopMatLoad(_880, blah, 0u, 16u, Layout);
-    coopMatStore(_880, blah, 0u, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _806;
+    coopMatLoad(_806, ssbo32.data, _549, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _815;
+    coopMatLoad(_815, ssbo32.data, _549, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _824;
+    coopMatLoad(_824, ssbo32.data, _549, 16u, Layout);
+    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseA> _857;
+    coopMatLoad(_857, ssbo32.data, _549, 16u, Layout);
+    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseB> _866;
+    coopMatLoad(_866, ssbo32.data, _549, 16u, Layout);
+    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseAccumulator> _875;
+    coopMatLoad(_875, ssbo32.data, _549, 16u, Layout);
+    coopmat<int, Scope, Rows, Columns, gl_MatrixUseA> _895;
+    coopMatLoad(_895, ssbo32.data, _549, 16u, Layout);
+    coopmat<int, Scope, Rows, Columns, gl_MatrixUseB> _904;
+    coopMatLoad(_904, ssbo32.data, _549, 16u, Layout);
+    coopmat<int, Scope, Rows, Columns, gl_MatrixUseAccumulator> _913;
+    coopMatLoad(_913, ssbo32.data, _549, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _923;
+    coopMatLoad(_923, blah, 0u, 16u, Layout);
+    coopMatStore(_923, blah, 0u, 16u, Layout);
 }
 
diff --git a/reference/shaders/comp/cooperative-matrix.vk.nocompat.comp.vk b/reference/shaders/comp/cooperative-matrix.vk.nocompat.comp.vk
index 951cc84a8..82898e3c0 100644
--- a/reference/shaders/comp/cooperative-matrix.vk.nocompat.comp.vk
+++ b/reference/shaders/comp/cooperative-matrix.vk.nocompat.comp.vk
@@ -30,57 +30,57 @@ shared uint blah[512];
 
 void loads_32()
 {
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _60;
-    coopMatLoad(_60, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _60;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _71;
+    coopMatLoad(_71, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _71;
     coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> A32 = tempArg;
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _79;
-    coopMatLoad(_79, ssbo16.data, 512u * gl_WorkGroupID.x, 32u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> tempArg_1 = _79;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _90;
+    coopMatLoad(_90, ssbo16.data, 512u * gl_WorkGroupID.x, 32u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> tempArg_1 = _90;
     A32 = tempArg_1;
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _92;
-    coopMatLoad(_92, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> tempArg_2 = _92;
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> B32 = tempArg_2;
     coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _103;
-    coopMatLoad(_103, ssbo16.data, 512u * gl_WorkGroupID.x, 32u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> tempArg_3 = _103;
+    coopMatLoad(_103, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> tempArg_2 = _103;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> B32 = tempArg_2;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _114;
+    coopMatLoad(_114, ssbo16.data, 512u * gl_WorkGroupID.x, 32u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> tempArg_3 = _114;
     B32 = tempArg_3;
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _116;
-    coopMatLoad(_116, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_4 = _116;
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> C32 = tempArg_4;
     coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _127;
-    coopMatLoad(_127, ssbo16.data, 512u * gl_WorkGroupID.x, 32u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_5 = _127;
+    coopMatLoad(_127, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_4 = _127;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> C32 = tempArg_4;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _138;
+    coopMatLoad(_138, ssbo16.data, 512u * gl_WorkGroupID.x, 32u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_5 = _138;
     C32 = tempArg_5;
 }
 
 void loads_16()
 {
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> _141;
-    coopMatLoad(_141, ssbo32.data, 128u * gl_WorkGroupID.x, 8u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _141;
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> A16 = tempArg;
     coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> _152;
-    coopMatLoad(_152, ssbo16.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> tempArg_1 = _152;
+    coopMatLoad(_152, ssbo32.data, 128u * gl_WorkGroupID.x, 8u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _152;
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> A16 = tempArg;
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> _163;
+    coopMatLoad(_163, ssbo16.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseA> tempArg_1 = _163;
     A16 = tempArg_1;
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> _164;
-    coopMatLoad(_164, ssbo32.data, 128u * gl_WorkGroupID.x, 8u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> tempArg_2 = _164;
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> B16 = tempArg_2;
     coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> _175;
-    coopMatLoad(_175, ssbo16.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> tempArg_3 = _175;
+    coopMatLoad(_175, ssbo32.data, 128u * gl_WorkGroupID.x, 8u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> tempArg_2 = _175;
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> B16 = tempArg_2;
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> _186;
+    coopMatLoad(_186, ssbo16.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseB> tempArg_3 = _186;
     B16 = tempArg_3;
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> _187;
-    coopMatLoad(_187, ssbo32.data, 128u * gl_WorkGroupID.x, 8u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_4 = _187;
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> C16 = tempArg_4;
     coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> _198;
-    coopMatLoad(_198, ssbo16.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_5 = _198;
+    coopMatLoad(_198, ssbo32.data, 128u * gl_WorkGroupID.x, 8u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_4 = _198;
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> C16 = tempArg_4;
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> _209;
+    coopMatLoad(_209, ssbo16.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<float16_t, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_5 = _209;
     C16 = tempArg_5;
 }
 
@@ -144,19 +144,27 @@ void scalar_construct()
     coopMatStore(A, ssbo32.data, 0u, 16u, Layout);
 }
 
+coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> coopmat_square(coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> a)
+{
+    return a * a;
+}
+
 void matmul()
 {
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _268;
-    coopMatLoad(_268, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _268;
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> A = tempArg;
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _279;
-    coopMatLoad(_279, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> tempArg_1 = _279;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> A;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> param = A;
+    A = coopmat_square(param);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _283;
+    coopMatLoad(_283, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _283;
+    A = tempArg;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> _293;
+    coopMatLoad(_293, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> tempArg_1 = _293;
     coopmat<float, Scope, Rows, Columns, gl_MatrixUseB> B = tempArg_1;
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _290;
-    coopMatLoad(_290, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_2 = _290;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> _304;
+    coopMatLoad(_304, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_2 = _304;
     coopmat<float, Scope, Rows, Columns, gl_MatrixUseAccumulator> C = tempArg_2;
     C = coopMatMulAdd(A, B, C, 0);
     C = coopMatMulAdd(A, B, C, 0);
@@ -165,47 +173,56 @@ void matmul()
 
 void matmul_uint()
 {
-    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseA> _315;
-    coopMatLoad(_315, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _315;
+    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseA> _329;
+    coopMatLoad(_329, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _329;
     coopmat<uint, Scope, Rows, Columns, gl_MatrixUseA> A = tempArg;
-    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseB> _328;
-    coopMatLoad(_328, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseB> tempArg_1 = _328;
+    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseB> _342;
+    coopMatLoad(_342, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseB> tempArg_1 = _342;
     coopmat<uint, Scope, Rows, Columns, gl_MatrixUseB> B = tempArg_1;
-    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseAccumulator> _341;
-    coopMatLoad(_341, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_2 = _341;
+    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseAccumulator> _355;
+    coopMatLoad(_355, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<uint, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_2 = _355;
     coopmat<uint, Scope, Rows, Columns, gl_MatrixUseAccumulator> C = tempArg_2;
     C = coopMatMulAdd(A, B, C, 31);
 }
 
 void matmul_int()
 {
-    coopmat<int, Scope, Rows, Columns, gl_MatrixUseA> _358;
-    coopMatLoad(_358, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<int, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _358;
+    coopmat<int, Scope, Rows, Columns, gl_MatrixUseA> _372;
+    coopMatLoad(_372, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<int, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _372;
     coopmat<int, Scope, Rows, Columns, gl_MatrixUseA> A = tempArg;
-    coopmat<int, Scope, Rows, Columns, gl_MatrixUseB> _371;
-    coopMatLoad(_371, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<int, Scope, Rows, Columns, gl_MatrixUseB> tempArg_1 = _371;
+    coopmat<int, Scope, Rows, Columns, gl_MatrixUseB> _385;
+    coopMatLoad(_385, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<int, Scope, Rows, Columns, gl_MatrixUseB> tempArg_1 = _385;
     coopmat<int, Scope, Rows, Columns, gl_MatrixUseB> B = tempArg_1;
-    coopmat<int, Scope, Rows, Columns, gl_MatrixUseAccumulator> _384;
-    coopMatLoad(_384, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
-    coopmat<int, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_2 = _384;
+    coopmat<int, Scope, Rows, Columns, gl_MatrixUseAccumulator> _398;
+    coopMatLoad(_398, ssbo32.data, 256u * gl_WorkGroupID.x, 16u, Layout);
+    coopmat<int, Scope, Rows, Columns, gl_MatrixUseAccumulator> tempArg_2 = _398;
     coopmat<int, Scope, Rows, Columns, gl_MatrixUseAccumulator> C = tempArg_2;
     C = coopMatMulAdd(A, B, C, 31);
 }
 
 void lds()
 {
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _482;
-    coopMatLoad(_482, blah, 0u, 16u, Layout);
-    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _482;
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> _496;
+    coopMatLoad(_496, blah, 0u, 16u, Layout);
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> tempArg = _496;
     coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> A = tempArg;
     coopMatStore(A, blah, 0u, 16u, Layout);
 }
 
+void array_of_coopmat()
+{
+    coopmat<float, Scope, Rows, Columns, gl_MatrixUseA> As[4];
+    for (int i = 0; i < 4; i++)
+    {
+        As[i] = coopmat<float, Scope, Rows, Columns, gl_MatrixUseA>(float(i));
+    }
+}
+
 void main()
 {
     loads_32();
@@ -220,5 +237,6 @@ void main()
     matmul_uint();
     matmul_int();
     lds();
+    array_of_coopmat();
 }
 
diff --git a/shaders/comp/cooperative-matrix.vk.nocompat.comp b/shaders/comp/cooperative-matrix.vk.nocompat.comp
index 6fd98e67c..a61e9e325 100644
--- a/shaders/comp/cooperative-matrix.vk.nocompat.comp
+++ b/shaders/comp/cooperative-matrix.vk.nocompat.comp
@@ -24,6 +24,11 @@ const int UseC = gl_MatrixUseAccumulator;
 layout(constant_id = 5) const int Layout = gl_CooperativeMatrixLayoutRowMajor;
 layout(constant_id = 6) const int Scope = gl_ScopeSubgroup;
 
+coopmat<float, Scope, Rows, Columns, UseA> coopmat_square(coopmat<float, Scope, Rows, Columns, UseA> a)
+{
+	return a * a;
+}
+
 void loads_32()
 {
 	coopmat<float, Scope, Rows, Columns, UseA> A32;
@@ -65,6 +70,7 @@ void matmul()
 	coopmat<float, Scope, Rows, Columns, UseA> A;
 	coopmat<float, Scope, Rows, Columns, UseB> B;
 	coopmat<float, Scope, Rows, Columns, UseC> C;
+	A = coopmat_square(A);
 	coopMatLoad(A, ssbo32.data, 256 * gl_WorkGroupID.x, 16, Layout);
 	coopMatLoad(B, ssbo32.data, 256 * gl_WorkGroupID.x, 16, Layout);
 	coopMatLoad(C, ssbo32.data, 256 * gl_WorkGroupID.x, 16, Layout);
@@ -164,6 +170,13 @@ void lds()
 	coopMatStore(A, blah, 0, 16, Layout);
 }
 
+void array_of_coopmat()
+{
+	coopmat<float, Scope, Rows, Columns, UseA> As[4];
+	for (int i = 0; i < 4; i++)
+		As[i] = coopmat<float, Scope, Rows, Columns, UseA>(float(i));
+}
+
 void main()
 {
 	loads_32();
@@ -178,4 +191,5 @@ void main()
 	matmul_uint();
 	matmul_int();
 	lds();
+	array_of_coopmat();
 }
diff --git a/spirv_glsl.cpp b/spirv_glsl.cpp
index b7d9c34f5..1ecb79bae 100644
--- a/spirv_glsl.cpp
+++ b/spirv_glsl.cpp
@@ -16329,13 +16329,17 @@ string CompilerGLSL::type_to_glsl(const SPIRType &type, uint32_t id)
 			require_extension_internal("GL_ARB_shader_atomic_counters");
 	}
 
-	if (type.op == spv::OpTypeCooperativeMatrixKHR)
+	const SPIRType *coop_type = &type;
+	while (is_pointer(*coop_type) || is_array(*coop_type))
+		coop_type = &get<SPIRType>(coop_type->parent_type);
+
+	if (coop_type->op == spv::OpTypeCooperativeMatrixKHR)
 	{
 		require_extension_internal("GL_KHR_cooperative_matrix");
 		if (!options.vulkan_semantics)
 			SPIRV_CROSS_THROW("Cooperative matrix only available in Vulkan.");
 		// GLSL doesn't support this as spec constant, which makes sense ...
-		uint32_t use_type = get<SPIRConstant>(type.cooperative.use_id).scalar();
+		uint32_t use_type = get<SPIRConstant>(coop_type->cooperative.use_id).scalar();
 
 		const char *use = nullptr;
 		switch (use_type)
@@ -16356,10 +16360,10 @@ string CompilerGLSL::type_to_glsl(const SPIRType &type, uint32_t id)
 			SPIRV_CROSS_THROW("Invalid matrix use.");
 		}
 
-		return join("coopmat<", type_to_glsl(get<SPIRType>(type.parent_type)), ", ",
-		            to_expression(type.cooperative.scope_id), ", ",
-		            to_expression(type.cooperative.rows_id), ", ",
-		            to_expression(type.cooperative.columns_id), ", ", use, ">");
+		return join("coopmat<", type_to_glsl(get<SPIRType>(coop_type->parent_type)), ", ",
+		            to_expression(coop_type->cooperative.scope_id), ", ",
+		            to_expression(coop_type->cooperative.rows_id), ", ",
+		            to_expression(coop_type->cooperative.columns_id), ", ", use, ">");
 	}
 
 	if (type.vecsize == 1 && type.columns == 1) // Scalar builtin